Last updated · 18 Nov 2026
Changelog
What's changed since SVG Lab went live. Newest at the top.
SVG Lab is generally available: the editor is free to use in any browser, no install and no signup, with paid plans for AI Generate and higher limits. It is actively improving: expect frequent additions. Bugs surface and get patched the same week. The version pill in the editor statusbar always tells you the current build.
18 Nov 2026
Version 1.71.0: your uploads in Claude, ChatGPT, the SVG Lab MCP and the API
- Your uploads, from your AI: SVG Lab in Claude and in ChatGPT, Claude Code, Codex, Cursor and any AI connected to the SVG Lab MCP can now list the files in your Uploads, place one in your design ("put my logo here"), read your design system section, whatever you named it, and save a design system board into it.
- The same in the API: four new /api/v1 routes do the same: list your uploads with download links, read your design system, place an upload in a project, and save a design system.
18 Nov 2026
Version 1.70.1: SVG Lab Agent is open
- SVG Lab Agent on Plus, Max and Ultra: the AI agent inside the editor is open to everyone on these plans. Open it from the AI menu, or with its button in Prototype mode and on the Chalk Board, and ask it to design screens, link a prototype or run a Chalk Board talk. It uses credits from your plan as it works. On other plans it shows you how to get it.
17 Nov 2026
Version 1.70.0: Auto-trace makes clean, editable vectors
- A new Auto-trace engine: logos, icons and flat illustrations now come back as a handful of clean shapes instead of hundreds of colour bands. Smooth colour becomes a real gradient, round parts become true circles, straight edges stay straight, and neighbouring shapes meet with no hairline gaps. On a flat leaf logo that used to make 1,163 shapes, it now makes 10.
- It takes a few seconds: most pictures trace in 2 to 10 seconds, and very detailed ones can take up to a minute. If a picture would take too long, Auto-trace uses the quick tracer instead and tells you.
- Edit every point: Place SVG now puts the trace on the canvas as ordinary shapes, grouped and sitting exactly over your picture. Press A (Direct Select) and click a shape to move its points and curve handles, or ungroup it to recolour each part. Very large traces (over 1,500 shapes) still arrive as one picture-like object so the editor stays fast.
- A readable Retry button: in the trace review, Retry now looks like the other buttons and sits next to Quality.
- Thanks: the engine is Vexel from Studi0Trace by Timothy Nice, used under its MIT licence.
17 Nov 2026
Version 1.69.0: your uploads in every project, and a place for your design system
- Uploads in every project: when you are signed in, the files you upload are kept in your account, with the names you gave them, and show in Uploads in every project. Double-click a name to rename it. Deleting one asks first, and copies already in a design stay.
- Design system section: Uploads has a second section for your design or brand system: guides, logos, colours and type, as PDF, PNG, SVG, Figma or token files (.json, .css). Rename it to anything you like, such as Brand system. Pictures in it place like any upload, and the rest stay together in one place in every project.
- Room for uploads: Free 50 MB, Starter 500 MB, Plus 1 GB, Max 2 GB, Ultra 5 GB, and 500 MB for each Team editor. One file can be up to 25 MB.
17 Nov 2026
Version 1.68.0: find any layer, and a Hand tool
- Search layers: a search bar at the top of Layers finds any layer by its name, its group's name or its words, on every artboard. A layer on another artboard shows that artboard's name, and clicking it takes you there.
- Rename a layer where it is: double-click a layer to type its new name, then press Enter (Esc keeps the old one). Undo brings the old name back. Names your AI gave stay until you change them.
- Pick one layer: clicking a layer in Layers now selects just that layer, even inside a group. Alt-click a layer to select its whole group.
- Hand tool: the Select button on the left now has a third choice, Hand. Click and drag to move around the canvas. Holding Space still works too.
- Choose files looks like a button: in Uploads it now has a light outline.
17 Nov 2026
Version 1.67.0: a canvas map, edit any part magnified, and many comments at once
- Canvas map: once a page has more than three artboards, a small map and a list of them sit on the canvas. Hover a name to see that artboard outlined, click it to jump straight there, drag it on the map to move the artboard, or drag names to reorder them. Each artboard shows as a small snapshot, and you can drag the map by its title to anywhere on the canvas or fold it away.
- Edit magnified: right-click a button, icon or any part and choose Edit magnified. It opens big on its own floating board beside its artboard, so small details are easy to change. Show nearby brings in the parts around it, and Save & exit puts your changes back in place, as one step you can undo.
- Many comments at once: leave comments on as many parts as you like, then send them all together to SVG Lab Agent, or to ChatGPT in the SVG Lab plugin, with one tap. Comments are kept with the project until they are done, so Claude, Codex or ChatGPT can read and fix them later.
- Comments stay where you made them: a comment left on a design no longer shows up on the Chalk Board.
16 Nov 2026
Version 1.66.3: SVG Lab is for teams too
- Clear about teams: our MCP page and the notes we give AI assistants now say plainly that teams use SVG Lab. Design by hand or with your AI, edit it yourself or ask the AI to change any part, and work on it together with shared team projects and free viewers. Before, they wrongly said SVG Lab was not for teams whose designs live in Figma.
- From design to code: they also say that after the design you get the UI as code, as a React app or as HTML, SwiftUI, Compose or Flutter, and that any artboard opens in Figma as layers.
15 Nov 2026
Version 1.66.2: help when a payment doesn't go through, and a cancellation email
- Try again, one tap: if you close the checkout and tell us your card or payment didn't work, SVG Lab now offers to open the same checkout again, and gives you our support address so we can sort it out with you.
- An email when you cancel: when you cancel your plan, billing@svglab.app emails you straight away with the date your plan ends and confirms you won't be charged again.
14 Nov 2026
Version 1.66.0: a new look, a better Chalk Board, and Chalk Board for your AI
- A new look: the canvas fills the window and every bar floats over it at one height, with the same corners and spacing, so buttons line up. The left rail shows icons (names on hover), and the selection bar sits at the bottom so the top of your canvas stays clear. Every big panel can dock to the side or float.
- Chalk Board for your AI: when you add SVG Lab to Claude, Cursor or another AI app with the address svglab.app/mcp, it can now sketch and explain on the Chalk Board: chalk words, arrows and boxes, and charts drawn by hand (pie, bar, line, 3D pie and bar, flow charts and mind maps).
- Chalk Board: it has its own right-click menu, and you can lock a selection so it stays put. Charts fit their space, show each label once and switch between 2D and 3D. Double-click a chart to see it full board, and press Esc to go back. Erasing puffs a little chalk dust, and an empty Chalk Board shows what it is for.
- Watch it work, piece by piece: when ChatGPT or Claude works in SVG Lab, the editor beside it follows the board being worked on and draws each new piece in as it arrives, instead of showing everything at once.
- All your screens at once in ChatGPT: after the first screen, SVG Lab adds every other planned screen in one step, laid out in rows.
- Prototype mode, clearer: the card is shorter with Play inside it, the link badge counts your links, a tap that changes how a screen looks says Changes to, and deleting a part that had links says which links went. Back deep in a flow no longer says first screen, and Copy link confirms with Link copied.
- Faster on big projects: grabbing a board group's border no longer freezes the tab, and zooming and switching pages are quicker. Board names stay readable when you zoom out.
- Smaller fixes: text keeps its size and place while you edit it, a group named after a board follows the board's new name, the save chip just says Saved (the time is in its tooltip), tooltips use the editor's own dark style, and a safety check stops a save from emptying a board by accident.
13 Nov 2026
Version 1.65.0: every app design comes with its design system
- A design system board in every app project: when you design app screens or a website with SVG Lab from ChatGPT, Claude or another AI app, your project now gets a board called "00 Design system". It shows every colour with its name and hex value, every text style with its font, size, line height and weight, the spacing, corner radius and sizes, and the buttons, fields and controls in your design's style. When the design has illustrations, it also shows the colours they are drawn in. Hand it to a developer as it is.
- Always up to date: SVG Lab builds the board from your design's own settings, so it matches your screens, and it rebuilds itself when the colours or fonts change. It is never treated as a screen: it stays out of Play, the screen list and the code export.
13 Nov 2026
Version 1.64.11: the editor opens on older iPhones again
- Older iPhones and iPads: the editor now opens on iOS versions before 16.4, including in-app browsers like TikTok's. Before, the page loaded but the editor never started.
- Resizing no longer stops working: dragging a resize handle right after the selected shape was removed (for example by an undo) no longer breaks the editor.
12 Nov 2026
Version 1.64.9: your credits, said plainly in ChatGPT
- Credits before the questions: when you start a design in ChatGPT, it first tells you how many SVG Lab credits you have left, out of how many.
- A warning at 90%: once 90% of your credits are used, ChatGPT tells you and links the Plans page, where you can see the plans and top-ups.
- Credits at the end: every design or change ends with how many credits it used and how many are left, and ChatGPT says so if the last credit is used up.
- Illustrations made the way you chose: SVG Lab's AI makes illustrations only when you picked it; if you chose hand drawn, they are drawn by hand.
- A cleaner side panel on phones: the SVG Lab panel beside ChatGPT no longer shows editing bars. It is for viewing your design and picking a part to ask about.
12 Nov 2026
Version 1.64.7: the email sign-in link works when you connect an AI app
- Email sign-in link fixed: when you connected ChatGPT or Claude to SVG Lab, or opened a team invite, the email link option stayed on "Finishing a quick security check" and never sent a link. It sends the link again. Google and GitHub sign-in were not affected.
11 Nov 2026
Version 1.64.5: a lighter projects list
- Your projects list loads light: the Profile page now lists your projects without downloading every drawing in them. A list that pulled every drawing at once is what stopped SVG Lab for about 40 minutes on 11 Nov (07:42 to 08:20 UTC).
9 Nov 2026
Version 1.64.1: faster starts in ChatGPT
- Your design starts sooner: in ChatGPT, SVG Lab opens with the first screen as soon as you have said what imagery you want. Any questions still open come right after, while the screen is already on the board.
- A full app is its core flows: about 12 to 20 screens, with menus, filters and dialogs added to the screen that opens them, instead of dozens of empty boards.
- Illustrations on big apps: an app with more than 24 screens now gets its illustrations on the screens they serve most. Before, the plan was refused and no illustration was drawn.
- Each photo once: a screen that repeats another screen's photo is held until it gets its own.
- Calmer empty boards: an empty board shows only the moving dots while it waits, without the grey shapes.
8 Nov 2026
Version 1.64.0: choose a plan and pay in two clicks
- A Choose button on every plan: the plans page now has a button next to each plan. It opens that plan on its own page with its price, its credits and what it includes, and a Pay button that takes you straight to checkout.
8 Nov 2026
Version 1.63.5: a heads up in your AI chat when credits run low
- Told before you run out: when you use SVG Lab from ChatGPT, Claude or another AI app, the chat now tells you once when you have 10 credits left (up to 2 app screens), and how to get more: a plan or a top-up from the Plans page. Before, the warning only came at the start of a task, so a design could stop halfway with no notice.
7 Nov 2026
Version 1.63.2: prices now say plus tax
- Prices exclude tax: every plan, Team seat and top-up price is in US dollars, plus any VAT or sales tax for your country. Paddle shows the tax and the total in your currency at checkout. The plan cards now say "Tax added at checkout" instead of "VAT included".
7 Nov 2026
Version 1.63.0: drag between places in Prototype mode
- Drag between places: pick a part in Prototype mode and choose Drag between places (under When swiped or dragged). A green copy of the part appears on the canvas: move it to where the part should rest, and add up to four places. In Play the part follows your finger, a flick carries it on, and it settles at the nearest place.
- Drop it on another part: drop a green copy on a cart, a folder or a bin and that part becomes the place. Let go near it and it is caught there; miss it and it springs back. Each place can go to a screen, change another part, or make the dragged part go away.
6 Nov 2026
Version 1.62.2: no blank side panel in ChatGPT
- The side panel shows your project: when ChatGPT checks on your project halfway through its work, the side panel now shows the design live instead of a blank white page. In the chat itself the check still takes no room.
5 Nov 2026
Version 1.62.1: a calmer placeholder while ChatGPT plans
- Drifting dots, no flashing light: in ChatGPT, an artboard waiting for its first piece now shows soft dots that slowly brighten and fade, like ChatGPT's own animation while it makes an image, instead of a bright light sweeping across it.
Version 1.62.0: SVG Lab in ChatGPT asks better, letters better and finishes cleanly
Hand lettering in ChatGPT, drawn in one go
- Hand-drawn lettering: when you ask SVG Lab in ChatGPT for letters drawn by hand, the whole lettering is now drawn in one go, like an illustration, with thick and thin brush strokes and tapered ends. Before, it was built letter by letter, which took many minutes and looked rough. It goes where your typed words were, and the typed words stay on their own artboard so you can still edit them. If a styled title was already there, the new lettering replaces it, so the words no longer show twice and no extra artboard is left in your project.
- Photo backgrounds: a photo that fills the whole post now goes behind your words straight away, instead of covering everything until it was moved back.
- Edits that fit: when you ask for a change, SVG Lab now has ChatGPT tidy what the change affects (for example the old framed copy of a photo you made the background) and tell you what else it changed.
- Dragging a photo: a cropped photo no longer disappears while you drag it.
- Advanced: the left bar's Details button is now called Advanced and sits above Uploads, with Layers at the bottom.
Posts with your photo as the background, and fonts picked for your brief
- Photo as the background: social posts, posters and store screenshots that use a photo now fill the whole design with it and put the words on a calm part of the picture. A framed photo appears only when the look calls for one or you ask.
- No more pale patches behind text: to keep words readable, SVG Lab changes the text colour, moves the words to a calmer spot or adds one soft shade from the edge, and tells you what else it changed.
- Fonts that fit: each look now picks the font pair that suits your brief (or the font you name) instead of the same serif every time.
- Looks you can picture: when SVG Lab offers lettering and poster looks, each one is described by what you will see ("letters painted with a brush: rough, streaky edges"), and the AI can show a small example of each before you choose. The recommended style now fits what you asked for, and there is a new elegant serif style for galleries, books, weddings and fashion.
- Real brush lettering: when you ask for a rough or dry brush, the letters themselves are painted that way, with streaks, broken edges and a loose hand, not smooth letters with lines on top.
Deleting an artboard deletes what is on it
- No more leftover shapes: when you delete an artboard, the shapes on it go too. They used to stay on the canvas outside every artboard. The dialog says how many shapes go, and Ctrl+Z brings the artboard and its shapes back in one step.
No more "(your copy)" artboards from a second tab
- Two editors on one project stay in step: with a project open in two places, for example SVG Lab beside ChatGPT and Edit in SVG Lab in another tab, the one you are not using no longer saves its own older copy over your changes. Moving, renaming or deleting an artboard in one tab no longer makes "(your copy)" artboards, brings deleted ones back or makes the artboard jump back.
- Saves no longer hang: a save that waited on a slow picture upload could show "Saving" for two minutes. It now goes ahead after about 12 seconds, and "Saving 1 of 2" counts only your artboards.
- Nothing lost by a reload: if you reload or close SVG Lab while a change is still on its way to your account, the browser now asks before you leave.
The side panel in ChatGPT says Done when ChatGPT is done
- No early "Design ready": while ChatGPT draws an illustration (it first makes reference pictures, which can take two minutes), the side panel stays live instead of saying the design is ready. The editor in the cloud stays open meanwhile, so the next step does not wait for it to start again.
- Done sooner: when ChatGPT finishes, the side panel stops saying "thinking" within 45 seconds, and at once when it ends with your PDF or PNG. Artboard names in the side panel no longer offer to be dragged: it shows the design, and your edits happen in Edit in SVG Lab.
- The side panel follows the AI: it shows the artboard ChatGPT is working on, even a new empty one, instead of staying on the first.
Clearer questions in ChatGPT
- Your poster title, your choice: for a poster, flyer, social post or invitation, SVG Lab in ChatGPT now asks how the big title should look: a clean font, or drawn by hand like an illustration.
- The cost is in the choice: when it asks how illustrations are made, the choice says what each costs (SVG Lab AI: 500 credits each; drawn by hand: no extra credits).
- No more stopping halfway: it asks more questions when something is unclear, and asks them right away instead of ending its turn by saying what it will ask next.
- Every new design asks first: a second design made soon after the first now gets its own questions too, instead of starting straight away and asking them as plain text in the chat.
- Choices you can tap, and see: when SVG Lab in ChatGPT offers looks or lettering styles, they come as choices you tap, each saying what it looks like in plain words, with an example picture when ChatGPT can search the web. No more tables to answer by typing a number.
- ChatGPT writes the questions: in ChatGPT Chat, the question card beside the chat now asks the questions ChatGPT wrote for your request, in its own words, instead of a fixed list. SVG Lab only adds whether you want pictures (if ChatGPT did not ask) and who draws the illustrations. Start in SVG Lab now says it is opening your project and counts the seconds.
- Work mode tip: in ChatGPT Chat, the question card now opens with a note that SVG Lab works best in ChatGPT's Work mode. You can switch, or carry on in Chat: the design is still made.
Lettering tells you when the words are too small
- A heads-up on small words: when you ask SVG Lab in ChatGPT to turn a small line into lettering, it still does it, and says the letters will be hard to read and offers a clean font instead. Curved small lines get the note too.
- Curves you can see: ChatGPT is now told how strong a curve must be to read as one, so an arched title is not bent so slightly that it looks like a mistake.
- Lettering takes the title's place: when ChatGPT turns a typed headline into lettering, the lettering replaces that headline in the same step, so you never see the words twice.
Easier selecting on posters
- Click a title anywhere on its letters: a click between the letters of a lettered title, or inside an O, now picks the title instead of the artboard behind it.
- Select things on a poster with a box: when a poster's background is a shape that fills the whole artboard, dragging on it now draws a selection box instead of sliding the background. A click still selects the background, and once it is selected you can drag it.
- Group names count artboards: a group of two artboards now reads "Neon Sign · 2 artboards" instead of "Neon Sign (2)", which looked like the name of a copy.
- Artboard group names: a click on a group's name selects the group, a double-click or right-click renames it, and names no longer overlap the artboards.
SVG Lab opens beside the chat right away
- No more waiting for the side panel: in ChatGPT's Work mode, SVG Lab now opens beside the chat as soon as it starts asking its questions, and your design appears there the moment you answer. Before, nothing showed until all the questions were answered.
- Links open cleanly: a project link from ChatGPT shows "Opening SVG Lab..." at once, instead of an empty editor with a Sign in button while the page loads.
Textures on any size of board, and a dry brush look for lettering
- Grain and paper on a whole cover or poster: ask the AI for film grain or a paper texture and it now adds it in one step, on any size of board. Before, a texture on a large board needed too many marks and failed.
- A dry brush look for lettering: ask for rough, brushed or painted lettering and the letters themselves get broken edges and dry streaks, in their own colours. It works on text, on letter shapes and on a whole group, and a drop shadow you already have keeps working.
- Easy to adjust: asking again on the same lettering or shape replaces the texture instead of stacking a second one, and one undo takes it away.
A slow check still reaches the AI
- No lost checks: when the AI looks at a heavy design and the look takes longer than SVG Lab waits, the finished look is now handed to the AI when it asks again, instead of being lost and rendered all over again. So it sees problems such as text that is hard to read before it hands the design over.
4 Nov 2026
Version 1.61.1: no more empty picture boxes when you choose lettering
- Lettering styles in plain words: when SVG Lab in ChatGPT asks which lettering style you want, it now names each style with one line about it. Before, it showed a picture for each style, and ChatGPT displayed those as empty boxes.
Version 1.61.0: 15 new kinds of lettering in ChatGPT
- More lettering looks: when you ask SVG Lab in ChatGPT for a poster, a sign or lettering art, it can now offer shiny chrome, neon signs, spray-paint letters, Art Deco gold, shop window and painted signs, chalkboard menus, grainy two-colour prints, wood-type posters, college team letters, tattoo flash, cut-out photo collages, pixel cross-stitch, warped type and words shaped into a picture.
- The right look for your brief: say "neon sign", "chalkboard menu", "letterpress poster" or "varsity lettering" and that look comes first.
Version 1.60.0: Claude connects SVG Lab for your whole team
- Claude auto-connect: on a Claude Team or Enterprise plan, your team's members can have SVG Lab connected the moment they log in to Claude, with no Connect button and no sign-in window. An owner or admin turns it on in the Team panel by pasting the Issuer URL from Claude's Managed authorization setup. How to set it up.
Version 1.59.2: the $3 AI pack is retired
- No more $3 unlock: an AI design made with the old free generation now exports once your account is on a paid plan. Until then you can still edit it, or delete it to export the rest of the project.
Version 1.59.0: AI Generate is for paid plans
- AI Generate now needs a paid plan: the free AI generation on the Free plan has ended. Every paid plan keeps AI Generate at 500 credits a generation from its monthly credits, and legacy Pro and Studio plans keep their monthly generations. The editor, the SVG Lab MCP and Auto-trace stay free to try on the Free plan.
- Designs you already made stay yours: an AI design already on your canvas stays editable, and a paid plan lets you export it.
3 Nov 2026
Version 1.58.10: the AI sees every export warning
- Warnings come first: when an export has something you should know, for example a font that could not be included, ChatGPT now always sees it and can tell you. Before, a long export pushed the warning past what ChatGPT reads, so it could stay hidden.
Version 1.58.9: exported text keeps its font
- The font travels with your file: an SVG exported by the AI now carries its font even when the small, letters-only font file could not be downloaded at that moment. SVG Lab uses the full font instead, so a logo or monogram looks the same on any computer, whether or not the font is installed there. Before, the file could name the font without containing it, and show a different font elsewhere.
Version 1.58.8: a font you name skips the letters question
- Named fonts, every time: when you name a font for a logo or monogram in ChatGPT, SVG Lab now puts it in your project's name, for example "Wedding monogram in Brygada 1918", so the question card no longer asks how the letters should be made. It reaches the ChatGPT plugin after OpenAI's next daily check of its tools.
Version 1.58.7: the live editor beside ChatGPT shows again
- Editor beside the chat: since the evening of 2 November, some ChatGPT users saw a broken page where the SVG Lab editor should be, because ChatGPT moved where it shows app cards. The editor now opens there again.
Version 1.58.6: lettering no longer piles up
- One lettering per spot: when the AI makes the same words again in the same place, the new lettering now replaces the earlier try instead of stacking on top of it. The same words somewhere else on the design stay, and undo brings the earlier one back.
Version 1.58.5: SVG Lab's question card in ChatGPT, steadier
- Start works the first time: when SVG Lab is still opening, the card now waits a few seconds and starts your design, instead of asking the questions again.
- No second project: going back to the card after you started shows your design, not the questions again.
- Your choices stay put: they are kept if the card reloads, and pressing Enter in a text box no longer starts the design.
Version 1.58.4: your letters, the way you chose them
- Drawn by hand means drawn by hand: when you choose letters drawn by hand for a logo in ChatGPT, the plugin now draws them, beside a version in a real font, instead of picking a font for you.
- No extra question when you named a font: the quick choices skip the letters question when you already said which font to use.
2 Nov 2026
Version 1.58.3: text to outlines works with Inter and other display fonts
- Outlines for engraving and cutting: exporting with text as outlines failed for large text in Inter, Bodoni Moda, DM Sans, Fraunces and a few other fonts that change shape with size ("Outline width differs from live text"). The outlines now follow the size you set, so the export works and the letters sit where you placed them.
Version 1.58.2: your fonts stay your fonts in logos
- You choose how the letters are made: for a logo or monogram, the ChatGPT plugin now asks once whether you want a real font, styled to fit, or letters drawn by hand. When you name a font, it uses that font and never redraws your words. For engraving or cutting, your text is turned into outlines in the exported file only.
Version 1.58.1: SVG Lab's question card shows in ChatGPT
- The card appears: SVG Lab's quick questions now show in their card in ChatGPT's Chat mode. In 1.58.0 ChatGPT pointed you to the card but did not show it.
Version 1.58.0: SVG Lab asks its own quick questions in ChatGPT
- Choices in the SVG Lab card: in ChatGPT's Chat mode, SVG Lab now shows its few quick choices (how much to make, the look, pictures, who draws the illustrations) in its own card. Press Start in SVG Lab and your design begins in your project, with the editor opening beside the chat. Before, answering ChatGPT's own question box could end in a flat picture instead of your design.
Version 1.57.2: ChatGPT Chat mode keeps your design in SVG Lab
- After any question: when ChatGPT asks you something mid-design (a look, who to draw, the event details), your answer now carries on in SVG Lab too, instead of turning into a flat picture.
Version 1.57.1: ChatGPT Chat mode makes your design in SVG Lab
- Your answers start the design: in ChatGPT's Chat mode, after you answered SVG Lab's quick questions, ChatGPT sometimes drew a flat picture with its own image tool instead. Now your answers go straight to SVG Lab, and you get an editable design in your project.
Version 1.57.0: any calculation by hand, smoother saves and faster switching in Prototype mode
- Any calculation, by hand: pick a text in Prototype mode and choose "Show a worked-out value": add, subtract, multiply, divide, percent, round, smaller or bigger, keep in a range, money, a list's total or row count, words joined, and If with and, or, unless. A calculation can use another one inside it, so "price times nights, as money" is one click on Show it here, and the panel says it in plain words before you make it.
- Saving without stalls: a big project's save is now worked out in small steps, so the editor keeps responding while it saves.
- Faster switching: going between Design and Prototype is faster on large projects, most of all right after an edit.
Version 1.56.2: any font name works in ChatGPT and Claude
- Ask for any font: when you name a font SVG Lab does not have, like Didot or Helvetica, your text, logo or design now uses the closest font SVG Lab has (Bodoni Moda for Didot, Inter for Helvetica) and keeps going. Before, only lettering did this, and the AI stopped to ask you what to do.
2 Nov 2026
Version 1.56.0: worked-out values, search on more lists and card morphs in Prototype mode
- Worked-out values you can build by hand: pick a text in Prototype mode and choose "Show a worked-out value": a number with its thousands marked (1,234.5), a time as a clock (1:30), how many letters were typed (a 12/140 counter), or words in capitals or small letters. It plays the same in Play, a share link, the website file and the React download.
- Search bars filter more lists: rows whose numbers differ in length, rows drawn right under the search bar, and a search bar with long words after its magnifier now filter as you type.
- Change all reaches copied screens: a change to a tab bar now goes to every screen that shows it, screens copied as "(your copy)" included, and one undo takes it all back.
- A View button morphs its card: Make it work now gives a "View" button inside a card a morph into that card's own page: the card's photo and name grow into place.
- Calendar days keep their look on their own date: in a calendar where you pick days, a day drawn faded, bold or coloured shows that look only on its own date. Other months show plain days, and a day picked in another month looks like any picked day.
- Downloads keep faded text faded: the React and website code downloads now keep a text's opacity, and an app that only shows remembered values, with no links, now runs in the React download too.
1 Nov 2026
Version 1.55.1: smoother logos in ChatGPT
- Lettering starts first time: when ChatGPT asked SVG Lab for lettering in a font SVG Lab does not have, the first try failed and it had to try again. It now uses the closest font SVG Lab has right away, for example Bodoni Moda for Didot.
- Clearer live view: the progress line beside your chat now names the part being made, like "Preparing an illustration" or "Planning the design".
Version 1.55.0: calm type designs get a lettering touch too
- Quiet, not plain: quote prints, script names, wedding signs and event posters made by the SVG Lab plugin in ChatGPT now always shape their main words in a way you can see from across the room, such as a word that changes colour where it crosses a shape, a swash under a name, paper grain inside the letters or a large ampersand between two names.
Version 1.54.2: AI Generate on a VPN, and the free picture in your AI app
- When a generation is cut off: if the connection drops while AI Generate is still working, SVG Lab now says so and suggests turning off a VPN or proxy before trying again, instead of a bare error.
- Why there is no free generation: an account made through a VPN or hosting network now sees that reason on the AI Generate button, instead of "Not enough credits".
- The free picture works the same in your AI app: your first AI picture made through the SVG Lab MCP or plugin is free to make and edit, like in the editor, and exporting the project it is in costs $3, which also adds 3 more pictures.
Version 1.54.0: the editor on your phone
- Your design, full screen: on a phone, or in a narrow ChatGPT panel, the editor now opens with your design filling the screen and fitted to it, under one slim bar with the project name, Undo, Redo and More. The desktop panels no longer pile up in one long column.
- Pinch and pan: pinch with two fingers to zoom, drag with one finger to move around, and tap anything to select it. Dragging something you selected moves it, so a stray swipe never moves your design.
- Big buttons at the bottom: Select, Text, Colour, Artboards and Export. Each opens a panel that slides up over the lower part of the screen: change words, pick a colour, jump between screens, or download SVG, PNG or PDF.
- Everything else under More: Layers, all settings, Share a link, Prototype, Chalk Board, sign in and Send feedback.
- Watching an AI design: the "ChatGPT is designing here" bar and its progress line now sit at the top and bottom edge of the screen, never over your design.
Version 1.53.0: type posters with real lettering in ChatGPT
- Designed letters, not typed words: when you ask the SVG Lab plugin in ChatGPT for a type poster, a quote print or calligraphy, the main words are now made as lettering. They can bend into an arch or a wave, lock lines together, carry an outline, a 3D edge or colour echoes, and hold a photo or a paper texture inside the letters.
- Cleaner letters: heavy capitals no longer crowd so close that letters run into each other.
Version 1.52.0: Friendlier emails, and a reminder when you leave a design half done
- A reminder for unfinished work: if you stop working on a design and don't come back for a few days, SVG Lab sends you one short email with a button that opens that same design, just as you left it.
- A new look: our emails now have a short note, one picture and one button, so the one thing that matters is easy to see.
- Credits in plain numbers: where an email mentions Starter, it now says what its 2,500 credits a month (up to 600 app screens, 500 auto-traces or 5 AI generations) can make.
Version 1.51.0: Simple graphics are drawn straight away
- Cards, curves and icons right away: when you ask SVG Lab in ChatGPT, Codex or Claude for simple graphics, like an app screen on a card, a pen curve with its anchor points, a cursor with a name tag, icon tiles, sparkles, squiggles or dots, it now draws them straight away instead of stopping. People, animals and other objects are still drawn with the same care as before.
Version 1.50.0: type posters, quote prints and calligraphy in ChatGPT
- Looks made for typography: ask the SVG Lab plugin in ChatGPT for a type poster, an event poster, a quote print or calligraphy and it offers looks made for them, such as Magazine quote poster, Words that follow a curve, Hand-written brush and pen lettering or Old-style badge and arch lettering, at poster, post and logo sizes.
- Words that look designed: the main words come as lettering with an outline, an edge or a curve, not plain typed text, and looks about ink or paper sit on a real paper or paint texture.
Version 1.49.0: Logos and headlines come as designed lettering
- Ask for a logo, get lettering: when you ask the SVG Lab plugin in ChatGPT or Codex for a logo, a wordmark, a badge or a headline, it now designs the letters themselves, with depth like colour bands, a 3D block or an outline, instead of typing your name in a font. You no longer need to find example pictures first.
- Pick a style from pictures: before it starts, the plugin shows SVG Lab's lettering styles as small pictures to choose from: retro script with colour bands, bubble letters, 3D block letters, outline and neon lines, brush script, arched badge, stacked poster and ribbon banner. The style you pick is the one it builds.
- Hand-built letters go through: letters the plugin builds one by one, touching and overlapping like a script, are no longer stopped as if they were a drawing.
Version 1.48.1: Edit in SVG Lab opens again
- Fixed: for a short while after version 1.48.0, Edit in SVG Lab in the ChatGPT plugin opened a page that was not found. It opens your design, signed in, again.
Version 1.48.0: Edit in SVG Lab opens your design signed in
- No sign-in on the way: in the SVG Lab plugin in ChatGPT, Edit in SVG Lab now opens your design in the browser already signed in to your SVG Lab account, on your phone too. If that browser is signed in to a different SVG Lab account, it asks before switching.
- No mixed messages: opening a link to a saved project while signed out now only asks you to sign in. It no longer also says your work isn't being saved.
Version 1.47.1: the right fonts in the editor beside your ChatGPT chat
- Fonts show as chosen: when ChatGPT set text in a font the SVG Lab editor beside the chat had not used yet, such as a script for a logo, the editor showed it in a plain stand-in font while the export looked right. It now loads the font as the work arrives, in that editor and in any other window open on the same project.
Version 1.47.0: designed headline lettering in ChatGPT and Codex
- Lettering, not just a fancier font: ask the SVG Lab plugin in ChatGPT or Codex for a stylish headline and it can now build real lettering: the words as letter shapes, stacked into a lockup or laid out freely, slanted or bent (arc, arch, wave, bulge, rise), with a solid 3D block, a flat shadow, an outline, echo outlines or a two-colour offset behind the letters. Every part is an editable shape in one named group, and one undo takes it all back.
- Eight new lettering faces: Lobster, Pacifico, Leckerli One, Great Vibes, Titan One, Bowlby One, Rubik Mono One and Monoton join the font list.
Version 1.46.0: App Store and Google Play screenshots in ChatGPT
- Looks made for store screenshots: ask the SVG Lab plugin in ChatGPT for App Store or Google Play screenshots and it offers looks made for them, such as Editorial and elegant, Huge bold headlines or Friendly illustrated, at the real store sizes, with your app's screens inside the phones.
- The look you pick is the one you get: a few looks share a name across kinds of design (a glassy look for apps and for desktop apps). The plugin now always uses the one for what you asked for.
Version 1.45.0: drag, pull, morph, search and live totals in Prototype mode
- Drag and pull: in Play and in shared prototypes, bottom sheets drag up to full height and down to close, and lists of alerts, messages or bookings pull down to refresh with a spinner.
- Swipe and scroll: cards swipe, rows reorder by their handle, rows wider than the screen scroll sideways, and long screens get a shrinking header, a reading bar, parts that stick and a back to top button.
- Morph to the next screen: a card can morph into its own page. The picture and words both screens share move and resize into place, the rest fades, and Back plays it in reverse. Make it work turns this on for cards only, and you can pair parts by hand in the panel.
- Totals that follow a pick: picking Express delivery ($16), a tip or a plan changes the total, the shipping line and the Pay button on later screens. A price per visit follows too: 60 min at Rs 2,600 makes 4 visits Rs 10,400, and "per 30-min visit" becomes "per 60-min visit".
- Change every copy at once: when you change how a tab or bar behaves on one screen, Prototype asks "Change all 7, or just this one?". One Ctrl+Z undoes it, and screens you copy pick up how their tab bar already behaves.
- See all, More and Read more: they can show more right on the screen, or open their page over the current one. Tap again to show less.
- Search that filters: typing in a search bar filters the list under it as you type, the count follows, and Enter opens the results screen.
- Buttons that went nowhere: in a row of action buttons, Request now opens what Send opens. Anything else that goes nowhere, or a screen you have not drawn yet (like a map for "Use my current location"), is listed with a prompt for your AI to draw it.
- Charts and calendars: a line chart's dot rides its line when the screen opens, with a pill that counts up to the value. Arrows next to a month or year step it, the calendar's days move with the month, and the days you pick keep their real dates.
- Hints off your design: in Play, link hints sit in a rail beside the phone, so they never cover your design.
- See every save: click the save status to see what was saved, its size, how long it took and why.
- Looking never saves: zooming, switching to Prototype or playing a project no longer saves it, so "saved 3 minutes ago" only moves when something really changed. After you play a prototype, your next edit saves only the screen you changed.
- The same everywhere: all of this plays the same in Play, share links, the website file and the React download.
31 Oct 2026
Version 1.43.2: ChatGPT and Codex: install the plugin, no developer mode
- One click in ChatGPT and Codex: the setup guide, the SVG Lab MCP page and the editor's Open in menu now send ChatGPT and Codex users to the SVG Lab plugin. Press + to install it, sign in to SVG Lab and approve. You no longer turn on developer mode, add a custom connector or paste an address. Claude, Cursor and other AI apps connect as before.
Version 1.43.1: no shimmer on empty artboards in the editor
- Empty artboards stay plain: a new, empty artboard in the SVG Lab editor no longer shows a moving grey placeholder. That shimmer is only meant for the editor inside ChatGPT, where it shows a screen the AI is about to fill.
Version 1.42.11: designs that arrive in Figma as you made them
- Figma-safe copies: Copy for Figma now writes the artboard the way Figma reads it: photo crops and rounded corners drawn into the photo, text with its font, weight and spacing, shadows written the way Figma writes them, and repeated parts written out in full. Anything Figma cannot keep, such as pattern fills, is changed to its closest look and you are told.
- Open in Figma in ChatGPT: the SVG Lab editor inside ChatGPT has a Figma button too, so you can take a screen from your chat straight to Figma.
- Blend modes kept in SVG exports: a plain SVG export used to drop blend modes; now it keeps them.
Version 1.42.10: paste from Figma
- Paste from Figma: in Figma, right-click a frame or shape, choose Copy/Paste as, then Copy as SVG, and paste into SVG Lab: it lands as editable shapes with its colours, selected, and one Undo takes it back. A plain Figma copy cannot be read outside Figma, so instead of pasting something old SVG Lab now shows how to copy it as SVG.
Version 1.42.9: share a design, not only a prototype
- Share in Design mode: the Share button now shows in Design mode too. Anyone with the link can view your design and leave feedback, no account needed. Screens you linked play like a prototype; screens with no links show as a slideshow, one at a time.
Version 1.42.8: Open in Figma, ChatGPT, Claude, Codex or Cursor
- Open in: a new Open in button at the top of the editor. Figma copies the artboard you are on and opens a new Figma file, so you press Ctrl+V there and it lands as layers. Below it, ChatGPT, Claude, Codex, Cursor and any other AI app can be set up to design here, each in a click or a few short steps.
Version 1.42.7: watch the design come to life, then Edit in SVG Lab
- A live bar on the editor beside ChatGPT: while the AI works, the top of the editor says what it is doing right now, such as "Laying out a screen: Home", with a moving line, while the screens change on the canvas below.
- Edit in SVG Lab: when the design is done, the bar says "Design ready" with the project name and how many screens it has, and a big Edit in SVG Lab button opens it in the full editor.
Version 1.42.6: two small error fixes
- Editing points no longer errors on a removed shape: if a shape was removed (by your AI or another tab) while its points were still on screen, clicking one of those points threw an error. Now the click does nothing.
- Quieter editor inside ChatGPT and Claude: the editor beside the chat no longer logs an error when the chat app blocks browser AI tools.
Version 1.42.5: drawings no longer get lost when Claude runs on its own
- Drawings finish in headless runs: when Claude Code draws illustrations through the SVG Lab MCP, it hands each one to a helper. In a run with nobody watching (for example
claude -p or the Agent SDK), a helper started in the background could be stopped when Claude finished its reply, and its drawing was lost. SVG Lab now tells Claude to keep its drawing helpers in the foreground and wait for every one, so each drawing lands in your project.
Version 1.42.3: the editor beside ChatGPT never falls behind
- It keeps up with the AI: the SVG Lab editor beside a ChatGPT chat could stop showing new screens partway through a design while the AI kept working, until you closed and opened the panel again. It now checks every few seconds while an AI is working and brings in anything new by itself.
Version 1.42.1: your AI uses fewer tokens per screen
- Shorter layout reports: when Claude, ChatGPT or Codex builds a screen through the SVG Lab MCP, the report SVG Lab sends back after each layout is now about 40% shorter. Each part of the screen is one short line (its name, position and size) instead of a block of fields. Your AI reads these reports again on every later step, so a whole app now uses fewer tokens and the chat fills up more slowly. Your AI can still ask for the full report when it needs every detail.
Version 1.42.0: looks made for desktop apps and platforms in ChatGPT
- Desktop apps get their own looks: ask the SVG Lab plugin in ChatGPT for a Mac or Windows app, a streaming or music platform, a live streaming site, a game launcher, a video editor, an audio plugin or a trading screen, and it offers looks made for that kind of app, such as Cinematic streaming, Audio gear with real knobs or Calm AI workspace, and builds at desktop size.
- Phone apps stay phone apps: a music player for iPhone, or a desktop website or dashboard, still gets the app and website looks as before.
Version 1.41.0: the SVG Lab editor opens beside the chat in the ChatGPT plugin
- The live editor, for everyone: when SVG Lab starts or opens a design from ChatGPT, the SVG Lab editor now opens beside the chat and follows the work as it happens. You can point at any part and add it to the chat. Before, the published plugin only showed a small picture card, and that card could say the live view was not available.
- It always finds your project: the editor beside the chat opens the right project even when ChatGPT hands it less information, and it moves on to the next project by itself.
Version 1.40.2: fewer, simpler questions in ChatGPT, and the credits each design used
- Tap to answer: on ChatGPT Free, the AI asked SVG Lab's questions as long written lists you had to read and answer in words. It now asks with choices you tap, a few at a time, in every ChatGPT plan.
- Only the questions that matter: it no longer asks where you want to watch, because the design opens beside the chat anyway. It asks how illustrations are made only after you choose illustrations or a mix, and never as a question with one choice.
- Plain words for the look: looks are named by what you will see, such as Bold and colourful or Glassy and premium, with a few words on why they suit your app, instead of design terms.
- Credits used, not just left: the last message of a design now says how many credits that design used, then how many you have left.
30 Oct 2026
Version 1.40.1: sign in with GitHub
- Continue with GitHub: you can now sign in to SVG Lab with your GitHub account, next to Google and the email link, in the editor, when your AI app (ChatGPT, Claude, Codex) connects to SVG Lab, and on team invitations.
Version 1.40.0: empty screens show they are on the way, and the AI asks SVG Lab's questions
- A shimmer on empty screens: a screen with nothing on it yet now shows soft grey layout shapes with a light sweep, so you can see the work is on its way while the AI plans. The first thing placed on it ends the shimmer for good. It is never part of your design or your exports.
- SVG Lab's questions, every time: in ChatGPT, the AI could ask a couple of questions of its own and start, skipping what SVG Lab needs to know. It now always asks SVG Lab's questions first: where the app runs, the full app or a set number of screens, the look, photos, illustrations or a mix, and where you want to watch.
- Choose how illustrations are made: SVG Lab's own AI illustrations (about 2 minutes each, 500 credits each, your first one free), or drawn by hand by your AI app from reference pictures (slower, no extra credits). You choose once at the start.
- Confirmation screens start clean: what a button opens (a "Message sent" or "Request received" screen) used to start as a copy of the screen behind it, and the AI wrote the new screen over the copy, so it looked broken for minutes. It now starts as a blank screen of its own.
- Faces stay in photos: a photo cropped to fit a frame now keeps more of its top, where faces are, instead of cutting a portrait off at the lips.
- A group stays on one screen: copying a screen for other phone sizes used to keep one group across the copies, so clicking a button picked the same button on every copy. Each copy now has its own groups, and a click picks only the screen you clicked on.
- The live view stays while the AI works: it no longer disappears in the middle of a run when the AI checks the project link, and the AI's "Thinking" tag no longer sits on top of your design.
- An honest finish: the AI no longer hands over the link while a screen is still empty, and its last message lists the screens, the hidden menus and sheets, anything left out, and your credits. It never shows a picture from another website as your result.
- Clearer picks in ChatGPT: a part you add to the chat now shows its own name and screen in the message box instead of "SVG Lab · 2", and the side panel opens fitted on the screen being worked on.
Version 1.39.0: photos stay put, screens never stack, and the AI asks first
- Photos stay in their frame: moving a photo with rounded corners, or the screen it sits on, used to leave its frame behind so the photo vanished. The frame now moves with it, and undo keeps the two together.
- No screen on top of another: when you drag a screen onto another one, or an AI app adds or moves one there, it slides to the nearest free space with its own artwork, and the empty starter board becomes the first screen instead of staying underneath.
- Questions first, as a form: in ChatGPT and Codex, SVG Lab now stops and asks before it builds, with clickable choices where the app offers them: a full app (every screen plus its menus, filters, profile and settings, all linked) or a set number of screens, the look, and photos or drawings. On ChatGPT Free it recommends photos.
- Delete a project from the chat: ask your AI app to delete an old project and it names it and asks you first; nothing is deleted until you say yes.
- A clear next step at the limit: when your projects or credits run out, the AI now tells you what to do: pick a plan in your SVG Lab account on svglab.app, or delete a project you no longer need.
- Cleaner details: a line break typed by the AI no longer shows up as a stray "\n" in your text, "Add to chat" names the part by its words and its screen, and the live view waits longer before saying the AI may have finished.
Version 1.38.1: the ChatGPT plugin starts your design again
- No more stopping before the start: since this morning's update, the SVG Lab plugin in ChatGPT could ask its questions and then say it was "unable to start the design" because a style step was not there yet. It now asks once and then builds.
- Questions for every new design (1.38.3): a second design started soon after the first now gets its own short questions too, instead of starting on guesses.
Version 1.38.0: a quicker, clearer checkout
- One-page checkout: choosing a plan or the AI pack now opens one checkout page with your details and payment together, with your country already filled in, so you go straight to paying.
- Prices in your currency: the checkout now shows and charges in your local currency where Paddle supports it, for example euros in Germany and pounds in the UK.
- The terms under every plan: each plan's button now says "Cancel any time · Secure checkout by Paddle · VAT included", so you know before you click.
- A heads-up on your last free export: on the Free plan, the export that uses your last free one this month now tells you so, with what Starter adds, instead of you finding out at the next export.
- Tell us what stopped you: if you close the checkout without paying, the one-tap question about why now reaches us every time, even when your browser blocks analytics. Usman reads every answer.
Version 1.37.2: a calmer opening from a project link
- Only the project you opened: when you open a project link from ChatGPT, Claude or another AI app, SVG Lab no longer shows your last project behind the loading screen while it fetches the one you asked for.
- Plain words while it opens: the loading screen now says what it is doing ("Opening your project", "Getting it from your account") instead of "Checking workspace" and "browser preview".
Version 1.36.0: the SVG Lab plugin in ChatGPT asks first, then builds
- Questions first, every time: in Chat, Work and Instant alike, SVG Lab now asks its few questions (where it runs, the screens, the look, photos or drawings, where to watch) in its first message, each with a recommended answer, and then builds without stopping to ask again.
- Allow it once, at the start: the first message tells you that ChatGPT will ask to let SVG Lab work in your account. Choose Always allow and a long design finishes while you are away, instead of waiting for a click halfway through.
- Named looks to pick from: SVG Lab offers a few design styles in plain words, with one recommended for your product, and builds every screen in the one you pick.
- Photos on ChatGPT Free: ChatGPT Free makes only a few images a day, so on that plan SVG Lab uses stock photos instead of drawn illustrations and never stops halfway at the image limit. On other plans, if image generation runs out mid-design, the empty spots get stock photos and you are told why.
- An honest finish: SVG Lab looks at each screen before calling it done, says what was made, anything not finished and why, gives the project link and the credits left, and lets go of the editor as soon as it is done.
Version 1.35.11: a steadier live view in ChatGPT
- Never a broken picture: if a picture of your screen is slow to arrive, the live view in ChatGPT keeps the last good one and tries again, instead of showing a broken image or an empty frame.
- Says when it is saved: when the AI is done or stops for a while, the live view says Saved or Done, tells you everything is saved in SVG Lab, and its button reads "Open in SVG Lab".
- Labels you can read: the label on the AI's cursor flips to stay inside the live view near an edge, and text appears once instead of twice while it is being written.
- The thinking clock counts from the last step: "Thinking" now counts from the AI's most recent step, so it no longer shows minutes while the AI is busy.
Version 1.35.9: your AI app reads your limits right
- Minutes and projects in plain words: when ChatGPT, Claude or another AI app checks your account, it now gets "25 of 30 design minutes used today, 5 left" and "3 of 3 projects used, 0 left", so it no longer shows "5 / 30" for the minutes you have left.
- App screens stay "up to": your AI app is told the app screens your credits make are a most, not a guess.
29 Oct 2026
Version 1.35.8: the live view in ChatGPT keeps up, and says when the AI is done
- Done when ChatGPT is done: when ChatGPT hands you the project link, the live view says Done and the editor gives the artboards back at once. Before, both kept saying "thinking" for a minute or two unless ChatGPT had paused the editor.
- Closer to the editor: the live view now takes a new picture of the screen after almost every change, instead of at most every 30 seconds.
- No more blank screen in the live view: when ChatGPT changed a screen right after its last picture and then only checked its work, the live view kept showing an empty artboard for minutes. Now the next step of any kind takes a fresh picture of that screen.
- No picture of an empty board: a new screen or drawing board shows up in the live view once something is on it. Until then the live view keeps the last screen and says what the AI is doing ("Studying a reference").
- Never stuck on Live: if the AI stops without handing over, the live view says Done about a minute and a half after its last step, instead of Live and "Thinking..." with a clock that kept counting.
- The steps no longer go backwards: once a screen has something on it, reading the design guides shows as "Checking the design guides", not as starting again.
- In the editor: after a minute with no changes it says "ChatGPT may have finished" instead of "ChatGPT is thinking", a tab opened mid-run lists what is already drawn instead of "Waiting for the first change", and when the AI lets go it says "ChatGPT finished. Its work is saved, and the artboards are yours to edit." Take over now says the artboards are yours.
Version 1.35.6: a Usage panel you can read at a glance
- One Credits section: the Usage panel shows each kind of credits once, with what is left and how many app screens it makes, for example "40 left (up to 9 app screens)". When you have credits in more than one place, their total comes first.
- Design time today: the minutes your AI app has used today sit in the same section, with what is left and what counts them (the time the editor your AI uses is open).
- Projects by name: the panel lists the projects your AI worked on, with their steps. Projects you deleted no longer show up, and internal tool names are gone.
- AI Generate on the Free plan: the panel now says what Free accounts get, your first generation free, and that paid plans use 500 credits a generation from their monthly credits.
- Your AI app knows your limits: when ChatGPT, Claude or another AI app checks your account, it now sees your projects left, design minutes left today and credits left, so it can tell you before a design runs into one.
Version 1.35.5: button labels where they belong in AI designs
- Labels inside their buttons: when ChatGPT, Claude or another AI builds a screen through the SVG Lab MCP and sends a button or a field as a box with its label after it, SVG Lab now puts the label inside the box. Before, the label could land under the button, white on a light page, so the button looked empty.
- Centred button labels: a label in a button the AI centred is centred, and one bold label in a button is centred too. Before, labels such as "Create account" and "Continue with Apple" could start at the button's left edge.
- Room around buttons: there is always at least 8 between a button and the hint, divider or button next to it, and text never touches the edge of the box it sits in.
- The AI sees what is wrong: the AI's layout check now names a label under its button, a label touching or off the centre of its button, text past the edge of the screen and tiny faint text, and every export tells the AI about any that are left, so it fixes them before it says the design is done.
Version 1.35.3: opening your work from ChatGPT
- Links open the right project: when you open a project link from ChatGPT or another AI app in a browser that is not signed in, SVG Lab asks you to sign in straight away and then opens that project. Before, it showed an empty canvas, asked you to sign up twice, and after sign-in opened your latest project instead.
- Unsaved changes save themselves: if a change never reached your account (a slow connection, or the page reloaded mid-save), SVG Lab now sends it on its own when nothing newer has been saved since. It only asks you when another device or app saved after that change. While a change is waiting, the save label says "1 change not saved" instead of "Saved".
- Sign out everywhere keeps your AI apps: the "Also disconnect my AI apps" box now starts unticked, so signing out never disconnects ChatGPT, Claude or another app unless you tick it.
- Nothing left on screen after signing out: the editor reloads to an empty canvas, so the next person on the same computer does not see your artboards.
28 Oct 2026
Version 1.35.0: pick dates on a calendar in Play
- Calendars you can tap: in Prototype mode, Make it work now finds a calendar (a month over its days). In Play, tap a day to pick it and tap it again to take it out. When you drew more than one day picked, several days can be picked at once; otherwise one.
- The words follow the dates: wherever your screens say the dates picked, they follow in their own format. "Nov 10, 11 & 12 · 3 visits" becomes "Nov 10, 12 & 14 · 3 visits", the button says "Continue · 1 visit" for one day, and "Fri, 16 Oct" becomes "Sat, 17 Oct" with the right weekday.
- Prices follow the count: a price that is the number of dates times a price per date follows too, so 4 visits at Rs 1,800 show "Rs 7,200" on the review screen.
- The calendar remembers: come back to it and the days you picked are still picked. Restart starts again from the days you drew. Days drawn greyed out, such as past or unavailable days, stay as they are.
- Share links and website files play all of this too.
Version 1.34.0: a shopping bag that adds up in Play
- Totals that follow the quantities: in Prototype mode, Make it work now finds a shopping bag. Tap + or − on a line and its quantity changes, and the subtotal, the item count, the total and the checkout button follow: "$260" becomes "$408" and "2 items" becomes "3 items".
- Your drawing stays as it is: a quantity drawn as one text ("− 1 +") works, and so does a number between a minus and a plus button. The + and − stay exactly where you drew them, and a quantity never goes below 1.
- Values worked out from others: the Values panel in Play (press V) shows the values your app works out, such as the subtotal, next to the ones you can change. The "Only if…" sentences in the panel can use them too, for example to show a free shipping note only when the bag is over $200.
- Prices read like a shop's: past 1,000 a price gets its comma ("$1,148"), and one item says "1 item".
- Share links and website files play all of this too.
Version 1.33.2: photo screens export from your AI every time
- Big screens export in one go: when Claude, ChatGPT or another AI exports a screen full of photos through the SVG Lab MCP, the export no longer fails with "Frozen export storage is unavailable". It used to run out of time on the busiest screens, such as a home feed.
- A whole app without waiting: an AI can now export every screen of an app one after another. Before, it had to wait up to 10 minutes after every fourth photo screen.
- The picture link comes first: each exported screen gets its picture link even when the export was slow, so no screen comes back without one.
27 Oct 2026
Version 1.33.0: teams that keep each other's work safe
- An artboard someone is editing is theirs: while a teammate works on an artboard, everyone else, owners included, can look at it but not change it. It opens again when they move to another artboard or stop editing for 3 minutes. No more surprise copies named "(your copy)".
- Viewers really only view: a viewer sees a View only note over the canvas, the tools that change things are off, and nothing they try is half done and then lost on refresh.
- Who is here: click a teammate's face to see their name, address, role and where they are, with Show me to jump there. Two people with the same name are told apart, and you never see yourself as someone else.
- A deleted project tells everyone: if a teammate deletes the project you have open, you hear it at once and can keep a copy in your own projects.
- The Team panel: removing someone is a clear Remove button, handing the team over is a list where you tick the new owner, and a double click on Invite no longer shows an error.
- People hear about changes: when an owner or admin changes your role, makes you the owner, or removes you from a team, SVG Lab emails you.
Version 1.32.0: onboarding from Claude swipes, tracking steps keep moving, and a quieter Check my app
- Onboarding that swipes: when Claude designs onboarding pages with page dots and hands your app back, the pages swipe one at a time and the dots follow. Next slides to the next page, and on the last page it goes where that page's own button goes (like Create account), so it never ends in a dead end.
- Fix: steps that move on by themselves (an order or request tracker) now keep moving after Claude redraws their screen. Before, a redraw left them standing still in Play. If yours stopped, one click on Make it work sets them going again.
- Check my app is quieter: it no longer lists rating stars, chat bubbles, step labels and other parts that only look tappable. It lists what is really missing.
- Make it work offers less noise: nothing on the Google and Apple sign-in sheets, which are already set up, and no list inside onboarding pages.
Version 1.31.0: tell us what stopped you
- One optional question: when you close a plan offer with Not now, or leave a checkout without paying, SVG Lab asks what stopped you. One tap answers it, and Skip closes it.
- Privacy: we now keep a session recording of how the editor is used (clicks, panels, scrolling) to find where people get stuck. Your drawings, every image, the code panel, all text on the page (so your layer, project and team names and your email) and everything you type are never recorded. The privacy policy has the details.
Version 1.30.1: connecting SVG Lab from Codex
- Fix: you can connect SVG Lab from Codex again. With newer versions of Codex, starting the connection ended on an error page before you could sign in. Now the SVG Lab sign-in page opens and Codex connects.
- Fix: after you press Approve, the sign-in page says "You're connected" only once your AI app has really picked up the sign-in. If it has not after about 20 seconds, the page says so and tells you what to do: in Codex, press Authenticate again or restart Codex; in other apps, start the connection again. Before, it said you were connected either way.
Version 1.30.0: connect your AI in one click
- Connect from the top of the MCP page: svglab.app/mcp-server now opens with a Connect your AI box. Pick Claude, ChatGPT, Codex, Claude Code, Cursor or VS Code and you get the shortest way in for that app, with the address and a first prompt to copy.
- One click for Claude, Cursor and VS Code: Add to Claude opens Claude with SVG Lab already filled in; Add to Cursor and Add to VS Code install it straight into the app. The setup guide has the same buttons.
Version 1.29.1: faster exports from Claude
- Faster exports: when Claude exports a screen with the SVG Lab MCP, the reply comes back about 2 seconds sooner on a big project. Before, it read the whole project again just to show the download card.
- Fix: Claude's last check before it hands your project back no longer counts copies and boards marked "safe to delete", the same as Check my app.
Version 1.29.0: screens from Claude arrive alive
- Calm entrances on Claude's screens: when Claude designs your app with the SVG Lab MCP and hands it back, each app screen it made that has no animation yet gets the Calm look of Feel alive: cards, titles and pictures fade in gently and numbers count up, so Play feels finished right away. Screens you or Claude already animated keep theirs, and Feel alive or Remove all animations changes them any time.
- Fix: Make it work now links a card or button that Feel alive animated. Before, it treated that part as already done, so it went nowhere in Play.
Version 1.28.1: held buttons and short app videos
- Fix: in Play and in share links, letting go of a button you hold (like an SOS button held for 3 seconds) no longer closes the menu it just opened. Before, the release landed on the menu and picked whatever was under it, with a mouse and on touch screens.
- Fix: a short app video no longer comes out empty. The newest Chrome (156) needs a second or three to get its video recorder ready, and pressing Stop before then said "Nothing was recorded". Stop now waits for it, shows Finishing, and gives you the video.
Version 1.28.0: the Team plan
- Teams for everyone: make a team in My Projects, invite people by email, and work in the same projects, with each person's name on the artboard they are editing.
- Team plan: $10 per editor seat a month, at least 2 seats. The team's owner pays; owners, admins and editors use a seat each, and viewers are free. Nobody invited needs a plan of their own.
- Each seat gets Starter: 2,500 credits a month (up to 600 app screens) for the SVG Lab MCP, AI Generate and Auto-trace, unlimited exports and 20 pages per document. Someone whose own plan is bigger keeps theirs. The team gets 100 cloud projects per seat.
- In the Team panel: the owner starts the plan, adds or removes seats (Paddle charges or credits the difference for the rest of the month) and can cancel it. A team without the plan keeps its projects open to everyone in it, but nobody new can join and no new project can be made there.
Version 1.27.0: Check my app names every dead end
- Every button that goes nowhere: Check my app lists each one with what it needs: a screen you already have that fits (a "Supplies" button can go to a "Shelters & resources" screen whose title says supplies), a screen still to draw, a page for one item of a list, or what happens after an action in a menu or sheet. Each row has a prompt to copy for your AI.
- Fewer false rows: buttons are named by the words you see. Copies, icon sheets and boards marked "safe to delete" are no longer listed, the pages of a pager count as reached through it, and a menu that opens when you hold a button counts as opened.
- Claude knows too: when Claude designs with the SVG Lab MCP, it hears about each dead end as it works, and before it hands the project back it sees what is still open and a map of every link.
Version 1.26.2: your drawing survives signing in
- Fix: signing in with Google no longer wipes what you drew while signed out. SVG Lab keeps the drawing through the sign-in and saves it to your account as a new project, and the "Leave site?" warning no longer appears for the sign-in itself.
Version 1.26.1: deleting the open project
- Fix: deleting the project you have open, from My Projects, no longer says it "was deleted on another device". SVG Lab now says it was deleted and opens your next project, or a blank canvas when it was your last one.
Version 1.26.0: the Prototype panel, at a glance
- A panel you can read at a glance: pick a part in Prototype mode and the panel shows its name, one line that says what it is, when it arrives and what a tap does, and two to four cards that fit it: a tab goes to its screen and lights up, code boxes fill as you type, a hold button fires only when held. Rest on a card to see what it does. Everything else waits under More, and the explanations sit behind a "?" beside the name.
- One animation on a whole row: "Arrives one by one" gives every tab of a tab bar, every option of a row or every row of a list the same entrance, 60 ms apart, in one click. Under More, a preset picked on one tab goes on all of them, one after another or all together, or on just that one. Parts picked together with Shift get the same choice.
- Screens from Claude work as they arrive: when Claude designs a screen with the SVG Lab MCP, its tab bar, switches, options, close buttons, steps, bars, counters and code boxes work right away, and Claude is told what was made. Undo takes the screen and its wiring back together.
- Fix: the Values panel in Play names a typed code by its label ("6-digit code") on every screen, never "A field" once the next screen shows.
Version 1.25.0: Check my app spots missing pages and steps
- Pages the dots promise: when a screen has page dots for more pages than you have drawn, Check my app now says so ("02 Onboarding has 3 page dots and 1 page: 2 pages are missing"), with a prompt to copy for your AI. When Claude designs with the SVG Lab MCP, it is told which pages to add, so Make it work can turn them into a pager with dots that follow.
- Steps the counter promises: a screen that says "Step 2 of 3" with no screen for step 3 is listed too, and Claude is told to design the next step or change the count.
- Fix: the "Copy a prompt for Claude" buttons in Check my app now work on the first click.
Version 1.24.0: sign-in that remembers you
- Signed in, in Play: your prototype now remembers whether someone is signed in. Allow on Google's sheet, Continue on Apple's and the right code all sign you in on the way to your home screen. Press V in Play to see Signed in in the Values panel. Share links, the website file and the React download play it the same way.
- Sign out that signs you out: Make it work turns your Sign out or Log out button into one that signs you out and goes back to your start screen, with no way back into the app.
- Pick the code that works: in Prototype mode, pick the code boxes or the Verify button and type one code under "The code that works". In Play only that code goes on, and any other shows your Wrong code screen. Leave it empty and any code of the right length works.
- Sign-in for screens you drew yourself: Make it work has a new Sign-in row. It builds Google's and Apple's sign-in sheets for your Continue with Google and Continue with Apple buttons, links your Log in buttons to your sign-in screen, and makes it all play. If people can tap Log in but there is no sign-in screen yet, it gives you a prompt to paste to your AI so it can draw one.
Version 1.23.1: screens from Claude arrive with clear parts
- Parts named for what they are: when Claude designs a screen with the SVG Lab MCP, the back arrow is its own part named Back button, each tile is one part named after its words (Report, Shelters), tabs read Tab: Home, and page dots and code boxes are named and numbered. The screen's background is called Background.
- One click picks the right thing in Prototype mode: a click on the back arrow picks the arrow alone, not the title beside it, and a tile's icon and words are picked together, so linking a control takes one click.
26 Oct 2026
Version 1.23.0: sign-in screens that play
- Google and Apple buttons as they really look: with the SVG Lab MCP, Claude can now place a Continue with Google button (Google's own four-colour G on a light or dark button, the way Google's guidelines show it) and a Continue with Apple button (black on light pages, white on dark ones), plus email and phone buttons in your app's own style. The Apple button shows its words only: Apple allows its logo only in mock-ups of apps for Apple devices, so SVG Lab does not include it.
- The sign-in sheets, ready made: tap Continue with Google in your prototype and Google's account choice opens, then the screen that asks you to allow access. Continue with Apple opens Apple's sheet. Allow and Continue land on the screen people see after signing in, and Cancel and the close button close them. Claude does not have to draw them. On a wide screen they open as a window in the middle.
- A code check that feels real: in Play, Verify shows a moment of checking that moves on by itself to the screen people land on. A code that is too short shows the Wrong code screen instead. Claude sets this up as it links the screens.
- Fixes from a real run with Claude: the sign-in count goes down as each piece is made and linked; colour and type names written with dashes (text-muted) work like the ones with underscores; a spacer inside a section that grows now takes the room it is given; and linking a button a second time replaces the first link in the panel and Check my app too, not only in Play.
Version 1.22.4: Prototype mode, lists that grow and shrink neatly
- What is under a list follows it: a line added to your bag now pushes the heading and the total below it down, and a row taken out lets them move back up.
- Long lists scroll: when a saved restaurant makes the Saved screen longer than the phone, the screen scrolls like a tall screen and its tab bar stays put, so the new row is never stuck behind the tabs.
- The right picture: a row added for an item shows that item's own picture, or the one on the page drawn for it, and never another item's photo: with none, a soft grey stands in its place.
Version 1.22.3: API fixes from the first live calls
- The first call works: an API call on a fresh connection no longer comes back empty.
- Test key PNGs are watermarked: PNG files made with a test key now carry the sandbox watermark, like SVG files.
- The docs example fits: the badge example on the API docs centres its word on the circle.
Version 1.21.1: move the AI progress card
- The card that shows what your AI is doing (Claude is thinking, the artboards under way, the latest steps) now starts clear of the tool bar on the left, so you can read all of it. Drag it anywhere on the canvas and it stays there while the AI works.
Version 1.21.0: the SVG Lab API, in early access
- The SVG Lab API: agents and scripts can now use SVG Lab's Design Engine over plain HTTP with a key: start a project, run Design Engine operations, make SVG with AI Generate, place free placeholder photos with their photographer's credit kept in the file, and download the finished SVG, PNG, PDF or code. It runs the same operations as the SVG Lab MCP. The API docs have one full example and every error with what to change.
- API keys on your Account page: name a key, choose live or test, set a daily credit cap, copy it once, and revoke it at any time (it stops on its next call). Test keys are a free sandbox with 1,000 test credits a month and watermarked files. Partner keys let a business serve its own customers, each with its own cap and a pause switch. Keys are in early access for now: write to support@svglab.app for one.
- Failed calls cost nothing: a call that fails gives its credits back, and so does a failed AI Generate.
Version 1.20.0: Prototype mode on Free and Starter, and emails you can stop
- Prototype mode on Free: Play, share links, every behaviour and most animations stay free. Thirteen animations now come with Starter ($7 a month): Ripple and Like when tapped, Bounce in and the chart entrances (Draw, Grow, Follow, Fill), the loading loops (Shimmer, Sweep, Dots), and Typewriter, Letters and Count up. Their cards carry a Starter tag, and you can still rest on one to see it play.
- Emails from SVG Lab: we may now email you when you reach a free limit, the day after you first use the SVG Lab MCP, and with offers and feature news. Every email has an unsubscribe link, where you can stop offers and news only, or every email. The privacy policy says exactly what we send and to whom.
Version 1.19.1: the newest SVG Lab MCP in every app
- Cursor, VS Code, Windsurf and every other app: connecting to https://svglab.app/mcp now gives you the same SVG Lab MCP Claude users get, with all its tools, instead of an older version. ChatGPT and Codex keep their own version. Nothing to change on your side: the next time your app connects, it gets the new tools.
Version 1.19.0: the SVG Lab MCP designs the sign-in screens an app promises
- Check my app reads sign-in: when your app lets people sign up, or says "I already have an account", "Log in" or "Continue with Google", Check my app now says what signing in still needs: a sign-in screen, a screen for the code with its Verifying and Wrong code states, and the Google and Apple sheets. It reads how people sign up (a phone number and a code, an email and a code, or an email and a password) and expects sign-in to work the same way; with a password it also asks for Forgot password.
- One click or a prompt: when the screen a button should open already exists, the row offers that link to keep. Otherwise Copy a prompt for Claude lists every missing piece in order, each with the step that makes it.
- Claude hears it while it designs: with the SVG Lab MCP, the first reply on a screen that offers sign-in names the whole set and the next step, and each later reply says how much is left, until every piece is designed and linked.
Version 1.18.0: plans made for the editor
- A clear next step at a limit: when a free account uses its last export or auto-trace of the month, runs out of cloud projects, or adds a second page, the editor now says what Starter adds ($7 a month: unlimited exports, 100 cloud projects with up to 20 pages each, and 2,500 credits a month (up to 600 app screens) for AI Generate, Auto-trace and the SVG Lab MCP). Get Starter opens checkout in a new tab, so your drawing stays open.
- Plans page: the editor comes first. Starter is the recommended plan and leads with unlimited exports. Prices and credits have not changed.
Version 1.17.1: Prototype mode, fixes from the live checks
- A shake that says "not yet": a button that does not work yet, and anything set to "Otherwise it shakes", now shakes from side to side for a moment, a little less each time, instead of swinging like a bell. When your device asks for less motion it still moves, just a little. It shakes the same way in Play, share links, the website file and the React download. Wiggle, for icons, still rocks.
- Chips saved the wrong way round: in some older projects, chips drawn picked were saved as not picked, and the other way round, so Play's Values panel and any "only if" read them backwards. Check my app now finds them and fixes them in one click, with nothing changing on screen. One Ctrl+Z undoes it.
- Step labels stay put: a progress bar beside "Step 2 of 3" still fills to two thirds, but its words no longer count up from "Step 0 of 3" as it fills. Feel alive and Count up leave "Step 2 of 3", "Page 3 / 5" and "Question 4 of 10" alone too.
- Counter buttons named by their row: the minus and plus beside a number are now called "Rescue boats, minus" and "Rescue boats, plus" in the panel and in Play, not "Add" or "Minus 2".
Version 1.17.0: Prototype mode, lists that change
- Hearts that save: a heart or SAVE puts its item in your Saved or Wishlist list, made from that list's own row design, and a second tap takes it out. An item already saved is never added twice, and once you change one, every heart for that item follows.
- A bag that fills: Add to bag adds the item the page shows as a new line in your bag, with its name, price, picture and the choices you picked, and "2 items" says "3 items". The same item with the same choices is not added twice.
- Replies that join the chat: a reply chip like "We can see you" adds your own bubble at the end of the chat, with the newest time already in it. The bubble fits its words, earlier bubbles move up, and the chips and the message field stay put. Send adds what you typed.
- Moving a row to another list: "Yes, cancel reservation" takes the booking out of Upcoming and into Past as Cancelled, and the Bookings tab then shows your "No bookings" screen.
- More from one click: Make it work finds hearts, Add to bag, reply chips and these deeds on your screens, each with a ✕ to leave it out. It also makes dropdown pickers remember their pick, and a swipe on a photo viewer's picture steps through the photos.
- In the panel: a row shows its words by field, with a box to type the words for one it lacks and its tags (what a filter reads when no word of its own says it). Each text a list's page fills has a ✕ to keep it as drawn, and two items drawn in different designs can be made one list.
Version 1.16.0: Prototype mode, only if and otherwise
- Works only if: a button can wait for something first. Continue on a form can work only once an option is picked: a tap before that shakes it, and it can also say why, such as "Pick what is happening first". You pick what it waits for from three short lists: the group of options, a switch, a value your app remembers or a field, then what it should be.
- Shows only if: a note can show for one option only, on the same screen, with no copy of the screen. "Medical: we alert the nearest ambulance too" shows while Medical is picked and hides when another type is, keeping its room, from the moment the screen appears. A hidden part cannot be tapped or typed in, and a hidden field shows nothing, not even while its words type in.
- Only if, on any line: any "Then also" line can happen only when something holds, and "+ Otherwise" says what happens when it does not, such as calling the fire brigade for Fire and 1122 for anything else.
- Make it work: two new kinds. Notes for one option finds words under a group that start with one option's words, and Buttons that wait finds Continue, Next or Send under a group with nothing picked yet.
- Check my app: it now says when a condition can never hold because its group or part was deleted, and when one always holds because an option is drawn picked.
- The same everywhere: it plays the same in Play, share links, the website file and the React download. The Values panel says when a tap did not work yet and lists the parts a rule hid.
Version 1.15.0: Prototype mode, the picked look slides along the row
- The look slides: in a row of options where the picked one has a pill, a line or a ring of its own (a segmented control, tabs with a line under the one you are on, chips with a pill behind the picked one, colour swatches with a ring), picking another moves that pill, line or ring to it and gives it that one's size. The options stay where they are, with their own words and icons. It plays in Play, share links, the website file and the React download, and lands at once when your device asks for less motion.
- Options with a box each: where every option has its own box (chips drawn each with their own fill), there is nothing to move: the old one's look fades out while the new one's fades in, in the same moment.
- Make it work offers it: rows whose picked one has its own pill, line or ring are listed as "Looks that slide", each with a ✕ to leave it out. One Ctrl+Z takes it all back. On a part of such a row, the "It becomes…" row offers "Slides along the row" next to "Gets picked", which fades the look in place.
- Segments that switch the screen: a segmented control whose segments each show the screen in another state (such as "Request details · Serious injury") now gets its looks from Make it work too, and keeps its links: the pill slides to the segment you tap while the screen switches in place. A copy kept as a screen of its own opens once the pill has landed.
Version 1.14.0: Prototype mode, counters and memory
- Counters: a number with a minus and a plus button beside it now counts. In Play, plus adds one and minus takes one away, and the number rolls up or down as it changes. It stops at its lowest and highest: "People with you" stays at 1 or more because it counts you, and "3 available" stops the boats at 3. Counters on a sheet keep the sheet open while you count.
- Your app remembers: a value set on one screen shows on every screen that shows it. Write its name in double braces in any text, such as {{People with you}}, and it shows the value as it is now. Back keeps what you set; Restart starts every value again.
- The Values panel: a new Values button in Play (or the V key) opens a panel beside the phone. It lists what your app remembers, what is picked, switched and typed, where steps and pages are, what is running, and what your last tap changed. Change a value there and the screen follows. It never shows in a recorded video. Add ?values=1 to a share link to see it there too.
- Make it work: a new kind, Counters, finds each number between a minus and a plus button, names it by the words beside it, and reads its highest from words such as "3 available". The panel says what each part does with it ("Then People with you goes up by 1"), and you can change where it starts and stops. Counters play the same in Play, share links, website files and the React app.
Version 1.13.0: Prototype mode, screens that switch in place
- One screen, several states: copies of a screen that show another option picked (No injury, Serious injury) are now that screen's states. In Play, share links, the website file and the React download, picking an option switches the screen in place: only what differs changes, a note fades in or out and what is below it slides.
- Nothing you set is lost: chips you turned on, words you typed, numbers a tap changed and where you scrolled stay as they were when the state switches, and Back brings the screen back in the state you left it.
- Every part works in every state: the screen's own links and taps work in all its states, so you set them once, on the screen. On a state's board, a part that is the screen's says so and shows its link as "from" the screen.
- Make it work finds them: Make it work like an app finds copies of a screen reached from an option on it, marks them as its states and links every option, the one drawn picked too, with a live preview of the switch and a ✕ to leave one out.
- No replays when you come back: a screen you come back to through its tab, a link or Back no longer plays its entrances again. Under "Again when you come back to it" the panel now says which: "Plays only the first time it opens" or "Plays every time it opens". A splash or loading screen that moves on by itself still plays when a link opens it again.
Version 1.12.2: Prototype mode, finishing touches from your videos
- Captions tap like their tiles: in Play, a tap on the words under a tile does what the tile does, also when the link sits on the icon alone. The same works on share links and in the website file.
- Back arrows tap on the arrow: a back arrow's tap area is the arrow and a little room around it, never the screen title or the icons beside it, even when they were drawn as one group.
- Plain names for icon buttons: a control drawn with no words is named for what it means. A bell is "Notifications" and a switch takes the words of its row, such as "Switch: Station flood alerts", in Play, the panel and Check app, instead of "Button" or "Toggle".
- Things that happen with no tap: when you link one part to another, you can now choose that it happens when the screen opens, 2 s after it opens, or when its entrances finish. The panel lists each one with what starts it and a ✕ to take it out.
- Steps from the same menu: on a list drawn as steps, that menu now offers Steps that move on: the current step gets the done mark and the next one lights up, every 4 s.
- See a screen change before you pick it: resting on Slide, Fade, Cover or Instant plays the link that way in the phone beside the canvas.
- Fitted dialogs in React: a dialog, sheet or dropdown you fitted shows smaller in the React download too, as it does in Play. After "Show it as its own screen instead", the message now says so.
- Cards stay in place in the React and HTML downloads: on a screen whose parts overlap, a card that holds overlapping parts too was drawn lower than in your design, over what came after it. On a status screen, the card with Call and Message covered the list below it. It now sits where you drew it, as in Play.
- Sliders show their track: the ▶ on a slider lights the track while the knob moves along it.
- Underlines fit their words: an Underline shown before its font arrived is measured again when the font comes, so the line fits the words on a first visit too, in Play, share links, the website file and the React download.
25 Oct 2026
Version 1.12.0: Prototype mode, lists you can read
- Every row opens its own page: tap any row of a list and the page you drew for one of them opens filled with that row's name, numbers and picture. Sara opens the sitter profile as Sara Ahmed, VIVIENNE opens the product page as The Vivienne at $192.
- Chips that filter: chips above a list now leave only the rows that say their words, the pins on a map follow their rows, and "18 helpers within 5 km" says how many show once the list changes.
- Filter sheets wait for their button: "Show 4 sitters" counts your picks as you make them, and the list changes only when you tap it. Close the sheet any other way and nothing changes.
- Sorting that sorts: Top rated, Nearest first, Price: low to high and Most recent put the rows in that order.
- Open a list for a word: category tiles, See all and search suggestions open their list for their own words. When nothing matches, your "No matches" screen shows, and Clear filters brings the rows back.
- Photos to browse: Next and Previous on a photo viewer step through the photos of the gallery that opens it, and "1 of 18" follows.
- In one click: Make it work finds all of this on your screens and lists each list, filter and sort with a ✕ beside it to leave one out. One Ctrl+Z takes it all back.
Version 1.11.1: the view stays put when the editor opens
- No jump right after opening: when screens appeared within the first moment after the editor opened (an AI building straight away, or a slow computer), the view could jump to the last screen made, so what you were about to click went off screen and a click picked nothing. The first fit now happens only while nothing is drawn yet.
Version 1.11.0: Prototype mode, bars that fill, loading, countdowns and numbers
- Fill: a new screen-open animation for progress, loading and capacity bars. The colour runs from the start of the track to the bar's value, like a thermometer. When the words beside it say "3 of 4" or "75%", it fills to exactly that, and the number counts up with it. Set it to fill to the end for a loading bar, and to play once or over and over.
- Loading: three new Always animations. Sweep runs a band along a bar that is still loading, Shimmer passes a soft light over a placeholder, and Dots pulses a row of dots one after another.
- Countdowns: Count up now counts down too, and can run live, a second at a time. "Resend in 0:42" counts down in minutes and seconds, then says "Resend code", and a tap starts it again. A dialog's "3, 2, 1" goes on to where its Send now button goes, unless Cancel closes it first.
- Numbers inside sentences: Count up is now offered on any text with a number in it, not only on a lone number, and only the number moves. Times left such as 0:42 count too.
- Its box first: a code box and its digit now arrive one after the other: the box comes in, then the digit types or counts in.
- Make it work: three new kinds. Bars fill to their values, a splash screen's loading bar fills to the end before the next screen opens, and countdowns count. They play the same way in Play, share links, website files and the React app.
Version 1.10.0: Prototype mode, type into fields and code boxes
- Code boxes that type: Make it work like an app now finds a row of code boxes, such as a 6-digit code from a text message. In Play you click the row and type: each digit fills the next box, the ring and the blinking caret move on, and Backspace takes them back. A pasted code fills every box, and a letter is left out of a code of digits.
- The button waits for the whole code: the button under a code row, such as Verify and continue, now goes on only once every box is filled. Before that it shakes, so nobody gets past with half a code.
- Every field takes typing: a field with a label above it now takes typing in Play even when it already holds a long sentence, such as a note for responders, and a taller box takes several lines.
- Typed words travel: Make it work can show what you type in a field wherever another screen shows the same words, and what you type stays when you go back, in Play, your share link, your website file and the React app.
- For recordings: pick a field and choose Type its words in when tapped: in Play its words then type themselves in, letter by letter.
Version 1.9.0: Prototype mode, Make it work now wires your screens together
- Links in one click: Make it work like an app now links buttons, rows and cards to the screen they lead to. "View profile" goes to your sitter's profile, a restaurant's row opens its page, Continue goes on to the next screen by its number. It only links a part when one screen clearly fits; when two could fit, it leaves the part for you.
- Menus too: a part that names a dropdown, sheet or dialog you drew for its screen now opens it, and so does a value with a dropdown arrow, such as an area or a sort order.
- See every link first: the window lists each link with its words and where it goes, and a ✕ beside each one leaves that one out. Back arrows still go back by themselves, so nothing links them.
- Fewer wrong offers: rows with a round mark are offered as options only, never as checkboxes too. A month's calendar, a row of figures and the faces inside a list are no longer offered as options, and a legend's colour key is no checkbox. A welcome screen with buttons no longer moves on by itself, and a loading screen moves on only to a screen it leads to.
- More found: switches drawn inside a row, Cancel, Got it and Keep buttons that close their sheet, the rows of a dropdown, and the rows of a screen named Choose now work in one click.
- Plain counts: after Make it work, the message counts one of a kind in the singular: "1 checkbox", "1 tab bar".
Version 1.8.0: Prototype mode, picked looks that carry everything, both ways
- A picked option looks like the ones drawn picked: tap an option in Play and it now takes their whole look: its check or badge comes in beside its words, the chip widens to fit it, and its colour, border and words change with it. Tap it again and all of that goes away. The same plays on share links, in the website file and in the React download.
- Both ways round: an option you drew picked loses its check, its colour and its extra width when it is unpicked, so on and off are always each other's opposite, whichever way you drew the part.
- Several at once keeps its group: the panel lists the options that go together in both modes, and clicking the card that is already in use opens or closes its settings instead of switching Several at once back to One at a time.
- Plain names you can try: the "It becomes" cards now say Gets picked, Flips on and off, Moves here and Stays, and resting on one plays it on your part and the ones beside it before you choose.
- Previews press first: the preview of a look (its card, the play button in its settings and the part's badge) now presses the part and then changes it, as a real tap does. An option that changes when tapped presses in Play too.
- It says what you drew: "Starts on, as drawn" or "Starts checked, as drawn" now follows your drawing, and "Start the other way" flips it.
- Lights up always shows: on a part with no picked look drawn, Lights up uses your accent colour, and when that would hide the part (a navy dot on a white card) it rings it instead. The link menu says so, and offers "Or it takes the look of" another part, so a ring can turn into the green tick beside it.
- Checkboxes: Make it work like an app now finds checkboxes. A tap on the box or on its words flips it: checked, the box fills and the tick draws itself in; unchecked, the tick fades out.
- The tap hint shows everywhere: when you tap where nothing happens in Play, every spot you can tap gets a light outline with a dark edge, so it shows on dark chips as well as white ones.
Version 1.7.6: Claude gets drawing steps it can follow
- Pictures in Claude: when Claude drew a picture straight onto a screen, SVG Lab's reply told it to make versions with image generation, which Claude does not have, so it worked around the step instead. The reply now gives Claude its own steps: find a reference, pick one, draw it by hand and place it, or use SVG Lab's vector model when you choose it.
Version 1.7.5: Prototype mode, entrances play the way you preview them
- Play matches the preview: what you see when you preview a screen's entrances on the canvas is now what plays in Play, on share links, in the website file and in the React download.
- Nothing shows before its turn: a text field whose words type in (a phone number, say) stays empty until its turn and then types in, instead of showing its words from the first moment. A counting number shows nothing until it starts, so no "0" waits on the screen. A code box whose digit types in comes in with its digit, never as an empty box first.
- Bars fill from their start: Grow, Draw and Slide in on a progress bar now fill it from its start to its value, like a thermometer, instead of fading the whole bar in. When an animation can only fade a part in (Draw on a part with no line, say), the message says so.
- Buttons and tabs move as one: Grow on a tab, a button or a chip grows it as one piece, its icon together with its label, and on a tab bar it goes tab by tab. Its tip no longer talks about bar charts: Draw, Grow and Follow describe what they do to the part you picked, and come first only for charts and bars.
- Reading order: a screen's parts arrive in reading order, top to bottom and left to right, with the back arrow first in its row. A part you give its first entrance takes its place in that order instead of going to the end, and the message says where ("It arrives 13th of 19").
- The order list shows the timing: each row says when it starts, and how long it waits first ("waits 1 s, starts at 1.5 s").
- The panel's first line: a part's panel now opens with when it arrives and with what, read from the same order list ("Arrives 4th of 6: Blur in · not linked"), so it never says None for a part that arrives.
Version 1.7.4: centred text stays centred in your downloads
- Centred and right-aligned text in exports: PDFs, PowerPoint files and image downloads moved every centred or right-aligned line a little to the left, so a letter inside a round logo sat against its left edge and a label could run off its badge. They now sit exactly where they are in the editor.
- Your AI's background colour sticks: when your AI set an artboard's background and its light or dark tone in the same step, the background was replaced by the theme colour. The background it asked for now stays.
- No watermarked references: the pictures SVG Lab shows Claude to draw from no longer include Unsplash+ previews, which carry a watermark and a paid licence.
Version 1.7.3: fewer false alarms and rebuilds when your AI designs
- No overflow reported from the artboard next door: a photo cropped to fill one artboard (a Story, say) was reported to your AI as artwork running past the edge of the artboard beside it, so it deleted or replaced things that were fine. Your AI's checks now only count what shows on the artboard they look at.
- Editing a wrapped headline keeps it wrapped: when your AI changed the words of a headline in a layout, the new words ran on one long line, off the slide and into the next one, and the next change to that layout brought the old words back. The new words now wrap in the same column and stay.
Version 1.7.2: fixes from a three-hour AI marketing run
- PowerPoint titles stay editable: a slide title or page number beside a photo cropped to fill its frame now exports as a text box you can edit. Before, the whole uncropped picture counted, so text next to it was baked into the slide image.
- Fit to selected artboard now frames the artboard holding what you have selected. Before, it framed the artboard you last switched to, so with a number on slide 5 selected it could jump to another slide.
- Tables with an empty corner: your AI can build a comparison table whose top left cell is empty. Before, the whole slide failed.
- Stickers and polaroids past the middle: a layout your AI places on the right half of an artboard, with no width given, now gets the room up to the right edge instead of failing.
- Clearer notes to your AI: it is told when it reuses a photo that is already on another artboard (so each piece can look different), when it edits a layout's text in a way that will not re-wrap, and when a photo address it sent was cut short. In Claude Code, it also tells you how to show the Browser pane when the live view opens hidden.
- Watching saves nothing: zooming in a tab that only watches your AI work no longer saves the view to the project.
24 Oct 2026
Version 1.7.1: watching your AI work never changes the project
- The live view only watches: a tab that has a project open while your AI designs in it (the browser beside Claude Code, or your own browser) no longer saves anything you did not change yourself. Before, it could save its own copy of the AI's artboards, undo the AI's latest work, make "(your copy)" artboards and bring back ones the AI had deleted. This held most for big projects: an AI's group of more than 24 artboards was only partly held, and tilted stickers and photos read differently in each browser.
- No made-up changes: your AI is no longer told you "resized" hundreds of texts when you did not. Text keeps the width it was saved with until its words or type change, so a browser with different fonts, or one whose fonts have not loaded yet, sees no change.
- Fewer save messages: because watching tabs no longer save, "One change to this project never reached your account" and "Another window saved changes to an artboard you were also editing" stop appearing while an AI works. A copy already waiting in your browser from before may be offered once; choose "Keep the copy on the server".
Version 1.7.0: Prototype mode, pages you swipe with dots that follow, and buttons you hold
- Pages with dots: an onboarding screen (or any screen with page dots) now swipes between its pages one at a time. The next page slides in under your finger, and the dots follow it: the long dot shrinks while the next one grows into it. A swipe past half a page, or a quick flick, turns the page; a shorter one springs back, and the first and last pages stay put. Tap a dot to go straight to its page. "Get started" shows on the last page only. It plays the same in Play, on share links, in the website file and in the React download.
- Make it work finds them: "Pages with dots" finds a row of page dots and the screens drawn as its pages (the same dots in the same place, or the same name), in the order their dots say. When the dots promise more pages than are drawn, it says how many are missing and copies a prompt for your AI to design them; it never draws a page itself. A star rating or a row of alike circles is left alone.
- Hold buttons: a button whose words say Hold ("SOS · Hold 3 s", "Press and hold") now works only when held that long, its ring filling as you hold it. Let go early and nothing happens; a single tap does nothing. Make it work finds them, and the panel's "When held" row says so, with Remove.
- Clearer swipe tiles: each swipe tile now says what it does before you pick it ("Throws it away to the left"), and on page dots the panel offers Pages first.
- Fixes: Slider on a part with no track under it waits until you pick the track, with one instruction at a time, and nothing is added if you press Esc. A track that is not under the knob is refused, so the knob never slides over the words beside it. The message and the panel name the same track, and the canvas shows it for a moment.
Version 1.6.0: Prototype mode, progress steps that move on by themselves
- Steps move on with no tap: a column of progress steps (a request's or an order's status, an onboarding checklist) now moves on by itself. Every 4 seconds after the screen opens, the current step gets the done mark (the check travels onto its ring) and the next step lights up, and it stops at the last step. It plays the same in Play, on share links, in the website file and in the React download.
- Make it work finds them: "Steps" finds a list with a mark beside each row, with the first rows drawn done, one drawn as the current step and the rest still to do, and makes it move on. Its preview shows your own list moving. One Ctrl+Z takes it back. Lists of actions, options and timelines are left alone.
- In the panel: with the screen in view, or a step picked, "Steps that move on" shows which step is current and how often they move on, with Remove.
- Fix: "Screens that move on" now goes on to the screen with the next number, never to a copy of the same screen such as "01 Splash (your copy)".
Version 1.5.2: Prototype mode, the React download plays as Play does
- Tap effects play to the end: in the downloaded React app, a tap effect such as Pop now plays all the way through before the next screen opens, as it does in Play. Before, the next screen came in when the effect was only a little over half done.
- Typed words come back: words typed into a field are still there when you come back to the screen with Back, and so are its switches, its picked options and how far it was scrolled. Fields you can type into in Play are real text fields in the download too.
- The same engine: the React app now runs the same code as Play and share links to decide what each tap, swipe, slider and screen that moves on by itself does, so all three behave the same. It is in src/prototype-engine.js in the download.
- Share links: new share links and website files are made in a newer format that the link's server checks field by field. Links you shared before keep working exactly as they do.
Version 1.5.1: Prototype mode, share links play as Play does
- One engine for both: Play, share links and the website file now run the same code to decide what a tap does, so a prototype behaves the same in all three. Two small differences went away: a menu item that goes to a tab's screen now adds a Back step on share links, as it did in Play, and a menu option that only picks something now closes its sheet in Play, as it did on share links.
- The tab you are on: when a tab bar is drawn as one part, tapping the tab of the screen you are on now scrolls that screen back to the top on share links and in the website file too, as it always did in Play.
Version 1.5.0: Prototype mode, screens that move on by themselves and words that keep their spaces
- A splash goes on by itself: a screen can now go on to another with no tap, once its animations finish or after 1, 2, 3 or 5 seconds. In Prototype mode, with nothing picked, set it under "After a while…", with how the next screen comes in. It plays the same in Play, on share links, in the website file and in the React code.
- Make it work does it for you: "Screens that move on" finds a start screen, or a splash, loading, sending or success screen, that goes nowhere, and makes it go on to the next screen once its animations finish. One Ctrl+Z takes it back.
- As in a real app: taps still work while it waits, and leaving it first stops it. The next screen takes its place, so Back never returns to a splash or a "sending" screen, and a screen you come back to with Back stays as you left it.
- No more dead ends: Play says when a screen goes nowhere and offers the next screen in one click. The Share dialog and Check app warn when your start screen goes nowhere, since everyone who opens the link would be stuck there, with the same fix. Press Update link afterwards so visitors get it.
- Words keep their spaces: on a first visit to a share link, "Word by word", "Letters" and "Typewriter" text could run together ("respondersnearby") when its font arrived late. The words now move into place as soon as the font arrives, in Play and on share links.
Version 1.4.0: Prototype mode, picking parts works the same way everywhere
- One click, one part: a click picks the part under the pointer, the whole button, tile, card or list row, never the screen's background. A double-click goes one level in, Esc goes back out, and a click on what is picked keeps it picked.
- A word or a number in a line: double-click it to give it an animation of its own. The line looks exactly the same.
- Select only, Ungroup: the right-click menu uses the words design tools use. Ungroup splits a part into parts of their own, for good, and its link stays on the piece you tap.
- Several parts at once: Shift-click, draw a box, or pick "Select all in this bar", then apply an animation to each or make them one part. Every part picked is outlined, with a count.
- Names you can read: parts are named after their words ("Shelters", "Page dot 2", "Checkbox: Share my live location…"), never "Button" for something that is not one, and no two parts on a screen share a name.
- Right-click shows what a part does, each animation and link with Remove, and the panel has its own right-click menu.
- With nothing picked, the panel names the screen you are looking at, with one line on what to do. Remove all now sits at the bottom.
- Menus and confirmations stay readable: menus open above the side panel and the toasts and move with their part; a confirmation (like Stop sharing) opens on its own dimmed layer; hover previews wait for the pointer to rest.
- The page never zooms away: a pinch over the canvas zooms the canvas. If the page is magnified anyway, Reset view puts it back in the same mode, screen and zoom. Signed out, it tells you how to pinch back instead, so nothing unsaved is lost.
- Play a tap you can see: Play a tap now shows its phone beside the panel instead of hidden behind it. On a small window the phone gets smaller so it fits.
- Fixes: screen names on the canvas no longer run into each other when zoomed out, a centred text no longer shifts when you click it zoomed out, and a click on a small button or icon that a link line runs across picks that button, not the line.
Version 1.3.0: Prototype mode, links you can see, pick and play
- Tab bars keep their tabs: picking several tabs together to give them an animation no longer gives every tab the link of one of them, and no tab loses its "Then also" rule. A link on a whole bar now says "Every tab in this bar goes to ...", with a button that sends each tab to its own screen. In Play and on share links, the tab of the screen you are on never takes you somewhere else.
- Link lines: lines run between the screens and leave each part on the side with no words or buttons in the way. Click one anywhere along it, over a screen too, to change it, play it or remove it, and rest on it to see where it starts and where it goes. A picked part shows only its own lines brightly, and "Then also" lines can be clicked the same way.
- Where does it go? Click a part's dot, or Goes to in the panel, for every screen and menu with a picture and a search. A screen's other states (its Medical tab, say) are listed under it. A new link that replaces another says so, with Undo.
- Drop the dot on a part: letting go of the dot on another part of the same screen asks what that part should do, and when (tapped or swiped), and can scroll the screen to it. Resting on a choice shows it first. What SVG Lab cannot do yet, such as a morph, is said plainly.
- Back arrows: one that goes back by itself shows a dashed "Back" beside its screen. Dragging it onto the screen it goes back to asks whether to keep going back.
- One Play: Play starts from the screen you are on and says which; "From the start" is its own button, and Ctrl+Enter and Ctrl+Shift+Enter do the same. Restart says which screen it restarts from, and Play's notes no longer cover the phone.
- Play a tap from anywhere: every link and rule has a ▶ in its panel row, its right-click menu and its line's menu. It plays the tap in a phone over the canvas, with its transition, and the canvas stays where it was.
- In Play: an option's tap area is its whole cell, at least 44 by 44. Show links (L) labels every spot you can tap with where it goes. A dialog as big as its screen can be fitted as a card, and a bottom sheet shows its handle.
- Clearer words: one name per setting ("How the next screen comes in"), "Lights up" rules say "lights up", and every Remove says what it removes. The Start screen list shows only your app's own screens.
Version 1.2.12: prototype share links play everything Play plays
- Share links kept only part of your prototype: a prototype opened from its share link now does everything it does in Play and in the downloaded website file. Before, a link left out what a tap changes besides its link ("Then also", going to another screen included), swipes and the Delete behind a row, sliders, picking one option of a set (the one picked before now goes back), screens that come up over the one before (Cover), tabs that add no Back step, a sheet that stays open until its own button closes it, fields you can type into on a menu or sheet, and fields drawn filled in (they now start with their words). A link you made before today keeps the old copy: open Share and press Update link once to fix it.
23 Oct 2026
Every plan says how many app screens it makes, and MCP early access ends: v1.2.3
- Credits in screens, everywhere: wherever SVG Lab quotes credits, it now says how many app screens they make with the SVG Lab MCP: 100 free credits (up to 24 app screens), Starter 2,500 credits a month (up to 600 app screens), Plus 7,400 credits a month (up to 1,800 app screens), Max 15,000 credits a month (up to 3,700 app screens) and Ultra 37,400 credits a month (up to 9,200 app screens). Top-ups: 5,000 credits (up to 1,200 app screens), 10,000 credits (up to 2,400 app screens) and 25,000 credits (up to 6,100 app screens).
- Shorter plan cards: each plan is one line about the SVG Lab MCP, then its limits.
- Early access to the SVG Lab MCP ends on Sunday 25 October, 12:00 noon Pacific time. No new requests are taken: anyone can connect now with the 100 free credits (up to 24 app screens) every account gets. The early access page opens the setup guide instead, with the address to add to your AI app. Accounts that had early access keep it until Sunday noon, then use their free credits, a plan or a top-up; their projects stay in My Projects and their AI app stays connected.
Connect your AI app from the AI panel: v1.2.9
- One place to connect: the "Connect your AI" button is gone from the top bar. Connect Claude, ChatGPT or Cursor from the AI panel in the left rail, which now says you start with 100 free credits (up to 24 app screens).
- The SVG Lab MCP in Usage: until your AI app has used the SVG Lab MCP, the Usage panel shows your free credits and how many app screens they make, with a Connect your AI button that opens the setup.
22 Oct 2026
The new editor layout, for everyone
- New layout: the editor opens in its new layout for everyone: tools in a rail on the left, a bar over the canvas for what you have selected, and panels that float over the canvas instead of squeezing it. Add ?ui=1 to the address to open the classic layout for a visit.
- The bar over the canvas stays: with nothing selected it now says so, names the artboard (click it to select the artboard) and offers Select all, instead of disappearing. Its buttons are never cut off on a laptop screen: the less used ones fold to icons with their names on hover.
- Right-click an artboard for its size and position, and to choose whether the artwork resizes and moves with it. The same switches in the size menu are now said plainly.
- Lock and unlock from the right-click menu: Lock this shape, and a right-click on a locked shape offers Unlock, so it can be freed from where it sits.
- Layers: a second click on the open artboard folds its layers away, and a right-click on an artboard row opens that artboard's menu.
- Uploads work like a library: uploaded files wait in the Uploads panel as tiles with a preview. Click one to place it, or drag it to where you want it on the artboard. A file dropped straight on the canvas is placed where you drop it and kept in Uploads too.
- 20 new icons, drawn for SVG Lab: a cloud with bumps on both sides, sun and cloud, rainbow, waves, hills, bird, fish, paw print, cactus, apple, balloon, rocket, light bulb, sparkles, crown, gem, key, trophy, tag and a smiley.
- Arrows follow your drag: the arrow shape now points from where you press to where you let go, in any direction. Shift snaps it to 45 degrees.
- AI Generate, redesigned: the window has the new look, and the result is shown large, in a frame the shape of the drawing, instead of a small picture in a wide white band.
- Chalk Board: only the board's own tools are on screen now, with Import (SVG or a picture), Export and the way back to the editor in a small bar of their own. One board is shown at a time, a Board button lists your boards to go to or delete one, and editor artwork pasted in near black turns chalk white so it shows on the board.
- Upgrade and AI are easy to find: Upgrade is back in the top bar for free accounts, and AI in the rail is highlighted.
- Usage shows the SVG Lab MCP: its credits left and the design minutes left today. In the MCP window, connected apps are listed once per app, with how many connections each has and one Disconnect for all of them.
- Fixed: a new page after Page 1 was called Page 3, and closing a page could land on the Chalk Board's boards. Rulers now run along the visible canvas instead of under the panels. The Details panel no longer scrolls sideways, its buttons are easier to hit, and its menus have the new look. Dropping an SVG onto the canvas no longer resizes the artboard to the file's size. Typing a new artboard width or height in the size menu and then clicking the canvas no longer throws an error and leaves the menu stuck open: the new size is applied and the menu closes.
Details in the left rail, and drawings that stay in one piece
- Details is in the left rail: one click opens colour, outline, size, position and every other setting of what you have selected. It is no longer repeated on the bar over the canvas.
- AI Generate results move as one: a generated drawing lands as one group, so dragging it never leaves pieces behind. Placing an SVG from Uploads works the same way.
- Auto-trace is in the AI panel again, under AI Generate.
- Duplicate makes its own group: a copy of a group no longer stays joined to the original.
- Fill an area colours the shape itself when you fill the inside of one shape, so deleting the shape removes its colour too, and filling it again replaces the colour instead of stacking a new one. A region walled in by several lines is replaced the same way, and goes when one of its lines is deleted.
- Layers lists each artboard's own things: a large group or a shape from another artboard no longer shows under the wrong one, and shapes on no artboard have their own row, Outside the artboards.
- The bar over the canvas never cuts off its buttons, even after you drag it somewhere else. Too long a bar scrolls with the mouse wheel.
- Chalk Board: a picture or SVG you import appears exactly as it is (no chalk texture, no colour changes) and stays one piece you cannot pick apart, on the board and in the exported file. Only what you draw gets the chalk look. The Board button becomes a dropdown once you have two or more boards.
- Fixed: imported SVG shapes in pure black, red, yellow or any colour with no blue in it came in invisible. They now appear as drawn. Send feedback now also says you can request a feature.
Buying a plan before you sign in
- Fixed: choosing a plan on the plans page while signed out now takes you to checkout for that plan as soon as you have signed in or signed up, with Google or with the email link. Before, you landed in the editor and had to find the plan again. Get Starter and Get Ultra also showed no sign-in window at all; they now ask you to sign in like the others.
One version number everywhere: v1.0.1
- Fixed: most pages of the site still showed an old version number (0.1.0) while the editor said 1.0.0. Every page and the editor now show the same version, and it goes up with every update.
SVG Lab for Business: v1.1.0
- New page, svglab.app/business: for companies that want SVG Lab, the everything-SVG platform, inside their own product: the Design Engine through its API and MCP, the editor inside their app, or their own brand as engine rules. Try the demos on the page, see what is live today and what we build with our first partners, and write the SVG Lab team a letter about what you're building.
Plans show what you get, and Start free: v1.2.0
- See what a plan makes: every plan now says how many app screens its credits design with the SVG Lab MCP.
- Start free: one click on the plans page or the MCP page creates your account, with its 100 free credits, and opens the setup guide for connecting your AI app.
- Choose a plan from the MCP page: the Choose Starter, Plus, Max and Ultra buttons take you straight to checkout for that plan, after sign-in if you are signed out, instead of to the plans page.
- Plans first: when you are signed out, the plans page opens on the plans themselves, on a phone too, and each plan leads with the SVG Lab MCP.
21 Oct 2026
The SVG Lab MCP in Claude: the live view ends on the finished screen
- Fixed: in claude.ai, the live view of a design could end on an empty space where the illustration goes, because the moment the drawing was placed came too soon after the previous picture. Placing a drawing on its screen now always updates the live view, and the drawing's own board leaves the strip once it is placed.
- Fewer approvals: Claude asks you before it uses each SVG Lab tool for the first time. Looking up fonts, components and the design's colours and type, and measuring a reference picture, now go through tools Claude already uses in every design, so a design asks you about four tools fewer.
- No scrolling to approve: claude.ai makes you scroll through a long step before you can allow it. In Claude Code, SVG Lab now sends a screen's layout, the colours and type, batches of shapes and a finished drawing by a short upload link, so the step you approve is a few lines long.
20 Oct 2026
The SVG Lab MCP in Claude: reference pictures it can draw from
- Fixed: the reference pictures SVG Lab found for an illustration were often watermarked previews from stock agencies, and Claude would not copy them, so illustrations went undrawn. SVG Lab now looks for Creative Commons pictures first, leaves stock-agency previews out, and tells Claude each picture's licence.
- New: at the start of a design, Claude tells you once how to stop claude.ai asking before every SVG Lab step: set SVG Lab's tool permissions to Always allow.
The SVG Lab MCP in Claude: duplicating a group and rebuilding a layout
- Fixed: when Claude duplicated a whole group (an illustration, a card), the copies landed inside the original group, so the next duplicate copied the earlier copies too. Each copy of a whole group is now its own group, with the original's name. Copying one shape of a group still keeps it in that group.
- Fixed: a gradient, pattern or filter Claude put on a layout's text or shape was lost whenever it changed that layout (moved it, resized it, edited another part). It now stays, unless the change sets that part's own colour.
- Checks: when Claude checks a design, it now hears about a divider or rule that runs into a drawing, which it used to call clean.
Prototype mode: the play button plays only what you applied
- Just the animation: the small play button beside a picked part now plays that part's own animation, right where it is. It no longer runs the link and jumps to the next screen. To see where a part goes, press Preview the link in the panel.
- Out of the way: the play button is now a small square just outside the part's corner, so it never covers the part or catches a click meant for it.
- The order is easy to find: the order your parts arrive in when a screen opens now sits at the top of the panel, as Order on and the screen's name. A picked part says where it arrives ("Arrives 2nd of 4") with a link straight to the list. After you preview a link, the panel stays on the screen you were working on, so its order and Play from here are still about that screen.
- Back arrows: the "Back by itself" label is gone from your design. Pick a back arrow and the panel says it goes back to the previous screen automatically, with a way to turn that off.
Fill an area works on rectangles and in big projects
- Fixed: clicking inside a plain rectangle or ellipse with Fill an area said "No enclosed region here", because only lines and paths counted as edges. Rectangles and ellipses now count too.
- Fixed: in a project with many screens, Fill an area could answer "Region too complex to trace" for a simple shape, because it looked at every shape on the page. It now looks only at the artboard you clicked.
- Fixed: imported SVGs drawn with arcs (rounded corners, circles, many icon sets) lost those curves and came in broken. Arcs now import as the curves they are, and rectangles with rounded corners keep their corners.
No false "never reached your account" warning after a slow save
- Fixed: on a slow connection, a save could reach your account after SVG Lab had stopped waiting for it. The next time you opened the project, a bar said one change "never reached your account" and that a newer copy had been saved since, when that newer copy was the same change. SVG Lab now checks what your account actually holds, clears the copy quietly when it is already there, and only asks you when something really differs.
The SVG Lab MCP in Claude: illustrations measured against their reference, then made your own
- Measured, not guessed: when Claude draws an illustration from the reference picture you picked, it now reads the picture in its drawing's own coordinates (zoomed crops with a grid, exact colours, and where each colour starts and stops), and after every round it lays the drawing over the picture to see exactly what is still off. It keeps going until the only differences left are the picture's soft edges.
- Then made your own: once the copy is exact, Claude gives the drawing a new pose by turning its parts at the joints and recolours it in your design's palette, so the illustration fits your design without losing the copy's quality. Every illustration goes through this step: the exact copy stays a study on its drawing board, and only the altered illustration goes onto your screens.
- Fixed: in Claude, fitting a drawing's artboard to its artwork used to stretch that artboard over every screen in the project. It now fits only the artwork on that artboard. Watermarks and stock logos on a reference picture are left out of the drawing and out of its measurements.
- Faster in Claude Code: Claude now downloads the reference picture and a small SVG Lab draw kit, measures and compares its drawing on your own computer in under a second per check, and sends SVG Lab only the finished drawing. Before, every check was a round trip to SVG Lab. The copy counts as finished once no shape or colour is off, so Claude no longer polishes the picture's soft edges for minutes. In claude.ai, which has no shell of its own, the same checks still run in SVG Lab.
19 Oct 2026
Big projects pan and zoom smoothly
- No more lag when you move around: scrolling, dragging the canvas and zooming in a large project (for example 20+ phone screens with photos) used to stutter at about 10 frames a second, because every small move made the browser restyle the whole drawing. Each move now costs about a fifth of the time, with no stutter.
Refreshing the page reopens the project you were working in
- Fixed: after you had opened a few large projects, refreshing SVG Lab could open an older project instead of the one you were working in, and kept opening that same one every time. A refresh now always reopens your current project.
- Each tab keeps its own project: with two projects open in two tabs, refreshing a tab reopens that tab's project. A new tab opens the project you used last.
- Lighter on your browser: SVG Lab keeps a local copy of only your recent projects, so your browser's storage no longer fills up.
Clicking, opening and switching to Prototype in big projects without the freeze
- Clicks select right away: in a large project (for example 20+ phone screens), clicking a shape could freeze the editor for one to two seconds before it was selected. It now takes a few hundredths of a second.
- Faster to open and to switch modes: opening a big project and going back from Prototype to Design no longer pause on the Layers list, which now only draws the rows you can see.
AI Generate: your free design goes straight onto the canvas
- Make it and edit it for free: your first AI Generate design now lands on the canvas as real shapes you can edit with every tool, at no cost.
- $3 to export it, with 3 more generations: when you want to export it, one $3 payment unlocks the design for SVG, PNG and PDF and adds 3 more generations (1,500 credits). No subscription.
- Accounts with credits are unchanged: plans and top-ups keep generating at 500 credits each, and the AI apps connected through the SVG Lab MCP keep their first free generation as before.
Prototype mode: swipes that go anywhere, Delete behind a row, sliders you tap
- Go to a screen from a rule: "Then also…" and every swipe can now go to another screen, open a menu or sheet, or go Back. Use "+ Go to…", or click a part on another screen and pick "Goes to" that screen. A swipe right on a row can open its details, a swipe up can open a sheet.
- Every swipe in the panel: Swipe left, right, up and down are all cards under "When swiped or dragged". Each swipe lists what it does, each with an ✕, and you can add another part that changes, a screen to go to, or the part itself going away.
- Delete behind a row, as on an iPhone: turn on "Delete button when swiped part way" on a row that swipes away. A short swipe stops on a red Delete; tap it to delete the row, tap anywhere else to close it. A long swipe still deletes at once.
- Tap a slider's track: in Play, a tap anywhere on the track glides the knob there, and you can keep dragging from that point. The words that show its value follow.
- Make it work finds more lists: rows with no box behind them (a shopping bag's items) and a single conversation or booking on a Messages or Bookings screen can now swipe away.
- ▶ on a part plays its rules: the ▶ on a part now also shows what else it changes on its screen (a list clearing, words changing, a row sliding out), and says what it does on other screens.
- Rules survive your AI rebuilding a screen: when the SVG Lab MCP rebuilds a screen, the rules that change its words or remove its rows now follow the new shapes, and it says which ones it could not match.
- Everywhere: Play, share links, the website file and the MCP all do the same.
Prototype mode: present a deck like a pro
- Speaker notes: each slide's panel has a box for what you want to say. Your AI can write them too.
- Presenter view: press S while a deck plays to see your notes, the next slide or build, a timer and the slide count beside the slide. On a share link, add ?present=1 to open it that way, and a second window of the same link (the one on the projector) follows your slides. Your notes stay private: a share link or website file carries them only if you tick Include speaker notes in Share. Without them, the presenter view of a link still shows the next slide, the timer and the count.
- Jump to any slide: type its number and press Enter, or press G (or Esc when presenting full screen) to see all slides and click one. G also lists every key.
- Play as slides, or as screens: a new switch in the Prototype panel plays any project as slides, or a deck as app screens. It is saved with the project and your share link follows it.
- Words on a card arrive with it: on a slide where everything comes in at once, the label or number drawn on a card now comes in with the card instead of waiting in place.
Prototype mode: fixes from a second test of real apps
- Pasting a copied screen: centred and right-aligned text now lands exactly 16 px from the original like everything else, instead of jumping sideways. Rotated text keeps its turn, and lines keep their spacing. The parts of the copy keep their taps and their "Then also" changes, so a copied screen plays like the original.
- Fields that already show a value: a box showing "Clifton" next to a search icon, or "Milo" under "Pet name", is now a field you can type in when you Play, in sheets too. It starts with its value, and a screen reader reads its label.
- The tab you are on: after Make it work, tapping the tab of the screen you are on scrolls that screen back to the top, in Play and on a shared link.
- Sheets close without a Back step: "Use this area" and other buttons that close a sheet back to the screen under it no longer make Back go to the same screen.
- Plain names everywhere: layers named "stack", "frame" or "container" are no longer shown as names. Menus, the Build prompt and the right-click menu say what the part is.
- Build prompt: a tab bar's tabs are listed left to right, even when one label sits a little higher, and the tab you are on says it scrolls the screen back to the top.
- Share link preview: the picture shown where you paste your link now uses your headings' own font (Fredoka and others) instead of a stand-in.
- Check my app is shorter: back arrows on screens no link reaches yet are one line ("31 back arrows will work once a link reaches their screens"), and a map pin or a bell in a top corner is no longer taken for a back arrow.
- Phone-width windows: the editor fits a 375 px window with no sideways scrolling, and the Design / Prototype switch is at the top of the panel.
- No save when nothing changed: setting a link to where it already goes, or picking the start screen again, no longer saves the project.
- Feel alive keeps photos under their captions: a photo inside a chat bubble animates with its bubble, so in Play it stays under its caption.
Prototype mode: the React code plays what Play plays
- Taps that change other parts: in the downloaded React app, a tap now does everything you set under "Then also": another option gets picked or flips, words change (a name, a size, a total), rows go away and what is below slides up, a sheet closes, the screen scrolls back to the top.
- Looks: switches flip, options and tabs get picked (the others go back), and Like turns only the heart or icon red, never the button or card behind it.
- Swipes and sliders: rows slide away when swiped, and a slider's knob drags along its track with its number changing beside it.
- Charts: Draw, Grow and Follow play in the app as they do in Play.
- Back shows a screen as you left it: its entrances do not play again, unless the screen asks to. A screen set to Cover comes up over the one before and slides back down.
- Decks: a deck downloads as slides that fit the window: arrows, Space or a click go on, each build comes in on its press, and slides change with their transitions.
- Older projects: the fixed success, warning and danger colours that projects had before 29 Sep no longer turn up in the code's colour names, so the download matches the SVG Lab MCP's export.
Prototype mode: the rest of Play in the React code
- Text effects: Typewriter, Word by word, Letters, Blur in, Count up and Underline now play in the downloaded React app when a screen opens, at the same moments as in Play. Words that come in with a deck's build play with that press.
- Pieces of one drawing: when a look moves, grows or fades one piece of a drawing (a switch's knob on its track, the dot of one tab among the tab bar's icons), that piece now does exactly that in the app, and each piece keeps its own colour.
- Ticks and lines copied from another option: the tick of a picked row or the line under the picked tab now shows in the app exactly where Play shows it, as its own small drawing.
- A card's background going away: a rule that removes only a card's background now takes away just that, and leaves its words where they are.
- Tabs: tapping between tabs no longer piles up Back steps in the app, as in Play.
- Decks: the React deck has a bar with Previous, the slide and build count, Next and Present (full screen), and a swipe on a phone goes on or back. Your speaker notes come along only when you have ticked Include speaker notes in Share; then S or Notes shows them with what comes next.
- Faster download: making the React code of a project with many behaviours is several times faster (Tably about 0.2 s instead of 1.3 to 1.8 s), with the same files.
Shared apps: Back and Restart over the phone
- Website file: the app you put on your website now has Back and Restart above the phone, lined up with its edges. Back shows once there is somewhere to go back to, as in Play.
- Share link: Back and Restart line up with the phone the same way.
- One button: when your app changed since you shared it, the Share window shows one Update link button instead of two.
The SVG Lab MCP: fixes from building a 15 screen app
- Moving a screen moves what is on it: when your AI moves a screen, its shapes, layouts, drawings and links go with it, the way dragging a board does in the editor. Before, the frame moved and everything on it stayed behind.
- Your AI can delete a screen: in Claude it can now delete a screen and everything on it in one step, and it says which links went with it and which screens linked to it. One undo brings it all back.
- No leftovers from a failed build: if your AI's screen layout stops halfway (for example a text with only spaces), nothing it drew is left behind, and a failed rebuild keeps the screen as it was.
- One undo after a rebuild: undoing your AI's rebuild of a screen now brings the old screen back. Before, it left the screen empty.
- Drawings keep their line weight in a smaller space: a drawing your AI places into a picture space now scales its outlines and dashes with it, so a big drawing in a small space no longer turns into dark blobs.
- Edits reach the right text: when your AI changes the size of a headline that sits after a group, the headline now changes, not the text after it.
- What it sees matches what it measures: the picture your AI looks at now draws Inter, Fraunces and other fonts with optical sizes exactly as the editor does, so text it measured to fit no longer looks too wide. Texts in Source Serif 4 can now be placed.
- The right button when words repeat: when a title and a button read the same words, your AI's menu or dialog now opens from the button, and it says which one it picked.
Prototype mode: the last rough edges in Play
- Sheets cover the tab bar: on a tall screen with a tab bar pinned at the bottom, a sheet or menu now sits over the bar, so a button in the sheet does what it says instead of opening the tab underneath. Play, share links and the website file.
- Delete behind a row in a sheet: a row in a sheet or menu that stops on a red Delete now shows the Delete right behind the row.
- A link and a "go" on one part: the panel now says a tap goes where its link goes, and offers "Go somewhere else instead", which replaces the link in one step (Ctrl+Z puts it back). Your AI follows the same rule.
- ▶ on sliders and Delete rows: the ▶ on a slider glides its knob to the top of its range, down and back, with its words following. On a row with a Delete behind it, it shows the Delete.
- Undo after your AI rebuilds a screen: one Ctrl+Z now brings back the screen as it was, and the rules that pointed at it.
- Words on a card arrive with it, in apps too: a label or number drawn on a card comes in with the card on app screens, and on a chart the labels fade in as the bars or line finish.
- An audience window from the editor: press S while a deck plays, then Audience window. A second window shows only the slides, follows you as you present, and never shows your notes. Drag it to the projector. Firefox and Safari leave full screen on the first Esc; the keys list now says so.
- Feel alive keeps the drawing order: a piece drawn between the layers of an animated drawing (a neck under a face) stays in its place in Play.
- A whole card swipes away: on a Saved or Bookings screen, a card with a photo, its words and its button is one row that goes in one swipe. A button and a hint box under it are no longer taken for a list, a setup screen such as "Enable reminders" offers no swipe, and the tab bar never slides up when a row goes.
- Tidier arrows: link arrows and the dotted "Then also" lines go round other screens through the gaps between them, and "Back by itself" sits beside its back arrow instead of on the status bar. Clicks still pick your design.
- Website pages keep Back: a link in a web page's nav bar now adds a Back step in Play, as a browser does.
18 Oct 2026
Chalk Board mode: explain things on a board, in chalk
- A second mode: the Editor | Chalk Board switch beside the SVG Lab logo turns the app into a chalk board: a dark board fills the screen, with one small toolbar. Good for teaching, explaining or talking through an idea on a screen recording. The same switch on the board's toolbar, or Esc, brings the editor back on the page you left.
- Everything in chalk: whatever you write or draw on the board has a chalk texture, with dry streaks and soft dusty edges, and an exported board keeps it. Your design pages are never touched.
- The tools you reach for: pen, arrow, line, rectangle, circle, hand-written text and an eraser you drag across what should go. Each has a key (P, A, L, R, O, T, E, or 1 to 8) and V moves things.
- Chalk colours: white, lime, yellow, pink and blue. Pick one before you draw; with Move picked, a colour repaints what you picked.
- Nine letterings: the Aa button picks how your text looks: Felt tip (the new default), Marker, Chalk, Print, Round, Quick note, Pen, Clean or Code. F steps to the next one. With Move picked, choosing one restyles the text you picked. Felt tip, Chalk, Print and Round are also new in the editor's font list.
- More boards: + Board starts a fresh one beside it, Page Up and Page Down (or a presentation clicker) move between them, Clear empties the one you are on and Ctrl+Z brings it back.
- A clean frame, never lost: H hides the toolbar for recording. A small tab stays at the top of the screen to bring it back, and H or Enter brings it back too. The Editor | Chalk Board switch also no longer spills out of the toolbar on wide screens.
- Kept apart from your design: boards live on their own Chalk Board page in your project, saved with it but never shown among your design pages. Export on the board's toolbar downloads the board you are on as an SVG.
The SVG Lab MCP in Claude: it plans with you, and you watch it on the canvas
- It asks first: in claude.ai, Claude Desktop and Claude Code, the SVG Lab MCP now asks a few questions before it builds anything (which platform, which screens, the look, photos or drawings, and where you want to watch), each with a recommended answer.
- Watch it in SVG Lab: it opens your project in the SVG Lab editor, in the browser beside your chat or in your own browser, so you see every screen made on SVG Lab's canvas. In claude.ai a live card in the chat follows the work too.
- Drawings you approve: it tells you how many illustrations go on each screen (none, one or several) and waits for your OK. Claude draws each one as SVG from a picture it finds on the web, or SVG Lab generates it for credits if you choose. In Claude Code each drawing gets its own helper, so they are made at the same time.
- Downloads as links: every exported screen comes back as a download link, which works in Claude Code too.
- Nothing to change on your side: Claude connected to svglab.app/mcp gets this automatically. ChatGPT, Codex and other AI apps keep what they have today.
Prototype mode: pick any piece of a part
- Pieces inside a part: double-click the number on a card, the name on a row or the icon in a button to pick just that piece and give it its own animation. The card keeps its own. The panel shows where you are ("Card › $12,480") and Esc goes back to the whole part. Its pieces are listed by name, so a tiny one can be picked by its name too.
- Shift-click keeps the rest: taking a piece out of a part, or adding one, no longer removes the animations from its other pieces.
- Right-click: on a part, a menu to pick just that piece or the whole part, link it to a screen, make it go back, remove its link or its animations, and its pieces by name. On a link arrow, Remove link comes first.
- The play pill moves out of the way: the green "opens ..." pill sits above or below the part, never on the design, and hides while your pointer is on the part.
- One record per part: a part's animations and links are now stored once, with its screen, so a card and the pieces inside it never disagree, and they survive your AI rebuilding the screen.
Prototype mode: you decide the order things arrive in
- When this screen opens: with nothing picked, the panel lists the screen's animated parts in the order they play. Drag a row, or use its arrows, to change the order, and choose whether each one starts with the one before it, just after it, or once it has finished. Your shared link and website file play the same order.
- Charts that come alive: three new animations for when a screen opens. Draw makes a chart line draw itself, Grow raises bars from their base one after another, and Follow sends a dot along the nearest line to its place. They work on any part.
- Screens stay as you left them: going Back no longer replays a screen's entrances, as in a real app. Tick "Again when you come back to it" under a screen's list if you want them to.
- A tap's animation finishes first: a button's Pop or Bounce now plays to its end before the next screen slides in.
- Tap the tab you are on: a link to the screen you are already on scrolls it smoothly back to the top. Any part can do this: pick it and choose "make it scroll to the top".
- Back arrows just work: the small back arrow in the top corner of a screen goes back in Play without being linked, and Check my app no longer lists it.
- Play from any screen: right-click a screen's name, or use Play from here in the panel. Ctrl+Enter (Cmd+Enter on a Mac) plays your app and stops it; with Shift it starts on the screen you picked.
- Try before you pick: point at an animation card to see it on your part without applying it; click to use it. The play button on a part now shows its entrance, its tap and its loop together.
Big projects stop freezing while they save
- Animations and links save in the background: on a project with dozens of screens, adding an animation, a link or pressing Undo used to freeze the editor for up to a minute while it saved. Now only the screen you changed is saved, in about a second, and you can keep working.
- Choosing a screen saves nothing: picking a screen in the Screens list, zooming or moving around no longer starts a save.
- Projects with photos open faster: photos now download several at a time, so a project with a dozen photos opens in seconds instead of half a minute, and they are no longer sent again on the first save after you open it.
- Smaller saves: layout pictures your AI placed are stored once with your photos instead of inside every save.
- No freeze while fonts arrive: opening a big project used to freeze the editor for several seconds when its fonts finished downloading, while every text was measured again one by one. They are now measured together, in a fraction of a second, with exactly the same result. Headings in Lora, Roboto and the other built-in fonts now also get their true size once their font arrives, so their selection box and alignment match the letters.
- Big projects open without a freeze: opening a project with dozens of screens used to lock the page for a second or more while every text was measured one by one. They are now measured together, so the project is on screen and usable almost at once, exactly as before.
Prototype mode: fixes from a full test of real apps
- Joining parts keeps their links: Shift-click a linked button and a label beside it, and the new part keeps the link. A row holding a piece with its own link can be changed again (its animation, its Remove, its arrow's Remove link), and the piece keeps its link. "Link removed" and similar lines only show when something changed, and a shape outside every screen no longer offers a link it cannot keep.
- Links never get in the way of your design: on a screen a click always picks the design, even where a link arrow passes over it. Arrows answer clicks in the space between screens, on a thinner line.
- Big cards and tab bars link like anything else: a card of many pieces is one part with its dot and its right-click menu, and dragging the dot of a piece you double-clicked, Shift-clicked or boxed links exactly that.
- Feel alive is calmer and smarter: it leaves the phone's status bar, the home indicator, slide numbers and drawing boards alone, never makes a tab bar slide in when a screen opens, and keeps an illustration whole and in its drawing order. Picking a second feel replaces the first (your own picks stay), and the dialog says which feel is on.
- Text effects: Typewriter, Word by word and Letters work on text with extra spaces ("ADD TO BAG · $148"). Count up counts the price, not the year beside it, keeps "01" as two digits, reads "€1.299,00" and leaves times, dates, versions, order numbers and "3/5" alone.
- Like turns only the heart red: it is offered where there is a heart or an icon, the words stay readable, and the pop stays inside the screen. Previews keep what is drawn above a part above it, as Play does.
- Check my app reads better: the status bar is never a "button that goes nowhere", icons are named from their layers ("Back arrow", "Close", "Send"), Show suggestions draws as many as it counts, "Keep all" keeps every suggestion at once, and clicking an item keeps your place in the list.
- Share links and website files use your fonts: every font of your project loads (Fredoka, Nunito and the rest), and the "SVG Lab" badge fits on a phone.
- App Store previews are 15 seconds or longer: a short take now holds its last screen long enough for App Store Connect. Stop sharing and Sign out everywhere ask in SVG Lab's own dialog, which works in every browser.
Prototype mode: make it work like an app, in one click
- Make it work: a new button next to Feel alive reads your screens and shows, on your own design, what it will make work: tab bars, switches, chips and other options, segmented controls and close buttons. Untick anything you want to leave as it is, then one click does the rest. Ctrl+Z takes it all back. The phone's status bar is never touched.
- Tab bars work like tab bars: each tab goes to its own screen, tabs swap at once instead of sliding, the tab of the screen you are on is shown picked, the bar stays pinned at the bottom of long screens, and Check my app lists a tab that goes nowhere. A bar your AI named "tab_bar", or any row of icons with labels at the foot of a screen, is now recognised.
- It becomes: click a part and choose what a tap does to it. Selected: it is picked and the ones beside it go back, in the look your design already gives the picked one (its colour, its tick, the line over the active tab). On / Off: a switch flips, its knob slides and its track lights up. Slides here: the pill of a segmented control slides to the one you tap.
- Options in menus: picking an option in a bottom sheet or dropdown moves its tick before the menu closes, and the menu remembers it next time.
- Everywhere the same: all of this plays the same in Play, in your share link and in the website file, and your AI can set it too.
Prototype mode: one tap, many changes
- Then also: under "When tapped", a part now lists what else its tap does, each said as a sentence ("Then the name on Send shows Liam Brown"). Click "+ Change another part", then click the part on any screen, or drag the new + dot onto it, and pick what happens to it: it gets picked or turns on, it flips, its words change, or it goes away. A dotted line on the canvas runs from the part you tap to each part it changes. Click ✕ on a line to take it out; Ctrl+Z works too.
- Words that change: a contact you tap shows its name on the next screen, a size you pick shows in the bag, the area you choose shows on the home screen. Only the words that change are replaced, so "Ivory · M" becomes "Ivory · S". Or type the new words yourself.
- Mark all as read, Clear, Delete all: the rows of the list fade away one after another and what is below slides up into their place, on any list.
- Pages that come up over the screen: a new way for the next screen to come in, Cover, for "More" and "See all": it slides up over the screen you are on, and its X or Back slides it back down. Pick it under "How it comes in" when a part goes to a screen.
- Sheets close the way they opened: a bottom sheet slides back down at the same pace it came up. A button in a sheet such as "Use this area" or "Apply" closes it and can change the words on the screen under it, and while a sheet has such a button, picking one of its options leaves it open.
- Make it work finds these too: Clear and Mark all read buttons by their list, a sheet's own button, and a choice your screens already show elsewhere.
- It stays the way you left it: picked options, changed words and cleared lists stay as they are when you go to another screen and come back, in Play, in your share link and in the website file. Restart puts everything back. Your AI can set all of it too.
Prototype mode: present your decks
- Decks play slide by slide: a pitch or sales deck no longer stops on its first slide. In Play, the right arrow, Space, Page Down or a click on the right half goes to the next slide; the left arrow, Page Up or the left half goes back; Home and End jump to the first and last; on a phone, swipe. A counter shows where you are. Slides need no links.
- Present full screen: press F (or Present) and the slide fills your screen, with black bars around it when the shapes differ. Esc leaves it.
- Builds on click: pick a slide and choose Build on click, and its parts with an entrance come in one press at a time, in the order of the slide's list. "With previous" and "After it ends" come along with the one before. Going back takes the last one away again, and a slide you return to shows complete.
- Slide transitions: None, Fade, Push, Slide and Zoom, each shown moving on its card. Set one for the whole deck or give a single slide its own; picking one plays it on the canvas.
- Made for decks: Feel alive gives each slide's headline, cards and pictures their entrance and the slides a matching transition, counts up only big amounts (never "01" or a slide number), and adds no taps. Check my deck no longer asks slides to link to each other, and Make it work explains that a deck already works.
- The same in your link: your share link and website file present the deck the same way, with its builds and transitions, and your AI can set transitions and builds too.
Share links: a real preview, replies and honest counts
- A preview card on X, WhatsApp and Facebook: a shared link now shows your app's name, a short description and a picture of its first screen, made when you create or update the link.
- Changed since you shared: after you edit your app, the Share dialog says so, with an Update link button next to it. The link keeps its address.
- Answer your notes: in Feedback you can reply to a note, mark it Done (its pin leaves the screen) or delete it for good. People with the link see your reply under their note.
- Views you can trust: a view is counted once per visit, so reloading the page no longer adds views, and the count in the dialog updates when you come back to it.
- A stopped link says so: after Stop sharing, the link shows a plain "This prototype is no longer shared" page.
Hand it over: a clearer Build prompt and the React code
- Build prompt with positions and words: it lists every part of each screen with where it sits, its size and its words, so a coding tool knows which button is which without reading a picture. Copy prompt no longer says a picture is attached when none is.
- Tabs change at once everywhere: the Build prompt and the React code now switch tabs instantly, as Play does.
- React code on every plan: Download React code in the Build prompt makes a React app of your screens, for every plan, up to 200 screens, in about a second.
- No waiting for a no: the screen pictures, the React code and the website file now check that you have an export left before they start, and count it only once your file is ready.
Prototype mode: swipe to delete, and sliders you can drag
- Swipe to delete: pick a row of a list and choose "Swipe left" under "When swiped or dragged". In Play the row follows your finger or mouse, slides out once you pass a third of it or flick it, and the rows below slide up into its place. A short swipe springs back. Swipe right works the same way.
- Sliders: pick the round knob of a slider and choose "Slider". Drag it along its track in Play: the coloured part of the track grows with it and the number beside it ("Distance: 40 km") changes as you drag. It stops at both ends, moves in the steps you set, and you can set where it starts and ends, its step, and how the number shows ("$#" shows $120).
- Scrolling still works: a long screen still scrolls up and down over the rows, a swipe to the right still goes back, and a bottom sheet still slides down, with a mouse or a finger.
- Make it work finds them: rows of a list that has a delete button or a Clear button beside it, and a knob on a long track with a number near it, each shown working on your own design before you tick it. The status bar is never touched, and Check my app no longer calls these parts dead buttons.
- Everywhere: swipes and sliders play the same in Play, in your share link and in the website file, and your AI can set them too.
Prototype mode: Play, previews and the editor, tidied up
- Back means back: pressing Back while a button shows its loading spinner stays on the screen you went back to; the app no longer jumps forward a moment later. Restart and Stop call the wait off too.
- Menus open over menus: a sheet opened from a sheet (dates from Filters) opens over it, and Back or Esc returns to the first one instead of closing both.
- Type into sheets and sign-in forms: search fields and text fields inside bottom sheets and dialogs take typing in Play, and an email field shown as "you@example.com" becomes a real email field next to the password.
- Tabs do not pile up Back steps: switching tabs replaces the screen, as in a real app, so Back leaves the tabs instead of walking through every tab you tapped.
- The keyboard stays in Play: Tab moves only between Play's own buttons and your app's tap areas, never to the editor behind it, and Stop puts the focus back where it was. Screen readers hear only the player.
- Previews you can trust: a tap's Speed shows in its preview (Slow is slower, Quick is quicker), and slow effects play to the end. Underline has a "Text colour" choice, picked unless you choose a colour, so what the panel shows is what draws. With "reduce motion" on in your system, a preview you click plays once and says that Play and your share link show the part still.
- Names a person would say: parts and menus named after raw layers ("icon_button", "icon icon chevron_down line") now show as "Button" or "Dropdown arrow" in the panel, on arrows, in Play, in your share link and the website file, and to screen readers. Your layers keep their names.
- See and steer your links: "Link to a screen…" lists your menus and sheets too. With nothing picked, the panel shows every link into the screen, and a click goes to it. "Hide link arrows" clears the canvas while the picked part's arrow still shows.
- Deleting a screen says what goes: the dialog and the status line say how many of its links and animations go with it and how many links to it will go nowhere. Ctrl+Z brings it all back.
- Locked parts: Check my app no longer links a part you locked (it could not be clicked anyway); it says to unlock it in Layers first.
- The editor stays put: at a 1024 px window the top bar fits and the page can no longer scroll sideways and get stuck. The Design / Prototype switch stays under your pointer, and switching keeps your zoom and the screen you were on. The zoom bar and floating panels no longer cover the part you picked or its play pill: the canvas moves it into the clear, and the link menu opens above the Screens panel.
- Your AI sessions: after you open another project, the editor checks only that project's AI sessions.
Your AI's screen moves and last edits now save
- Moved screens stay moved: when your AI moved a screen to a new place through the SVG Lab MCP, the save could be refused and the screen jumped back where it was. Moving a screen, clearing one, moving its contents to another or rebuilding it now saves, while a window with an out of date copy of your project is still stopped from saving over it.
- The last edit is not cut off: when your AI went quiet while a big save was still uploading, its editor could close before the save finished. It now waits for that save to land first.
17 Oct 2026
An old browser tab can no longer erase your artboards
- What went wrong: a project tab left open for a day and reopened by the browser showed an out-of-date copy, and when it saved, that copy replaced the real project: every artboard came back empty. This is fixed, and your account now refuses that kind of save on its own as well.
- Out-of-date copy, read-only: if your saved project takes too long to load, the editor shows your last copy with "Loading your account" in the save chip and does not save it. Saving starts once the real project is on screen; the load is tried again on its own.
- A save never deletes an artboard you still have: an artboard you cleared keeps its place in your account, and an artboard is only removed from your account after you deleted it in the editor. A save that would have removed artboards still on your screens is refused and the project is reloaded instead.
- Stored artwork is kept longer: drawings kept in storage are only tidied away a day after nothing points at them, not minutes.
16 Oct 2026
Prototype mode: make your screens work like a real app
- Pick how each part moves by watching it: switch to Prototype at the top of the right-hand panel, click a button, card, title or icon, and choose an animation from live previews: when it is tapped, when its screen opens, its text, or all the time. Adjust strength, speed and direction by eye.
- Link screens by dragging: drag the dot on a part to another screen. Menus, side menus, bottom sheets and dialogs open over a screen, with starters when your design has none.
- Make it feel alive in one click: watch three feels on your own screen (calm, snappy, playful) and pick one. Every button, card and title gets it. Remove all animations, or all links, takes them off again in one click, and Ctrl+Z brings them back.
- Play your app: tap through it from the start screen. Long screens and web pages scroll with their top and bottom bars pinned, switches flip, text fields take typing, buttons can show a loading spinner, swipes go forward and back, and every screen remembers what you did on it.
- Check my app: finds buttons that go nowhere, screens nobody can reach and links to deleted screens, and suggests the missing links. A tab bar linked as one piece is split so each tab goes to its own screen, in one click.
- Share a link: anyone can open it on their phone, no account needed, and leave notes everyone with the link can see. Links expire after 24 hours unless you pick 1 hour, 3, 7 or 30 days. A QR code opens it on your phone, and new notes show on the Share button.
- Put it on your website: download one HTML file that plays your app, animations and taps included.
- Make a video: record your app as an MP4 for X, or as an App Store preview at the size Apple asks for (Chrome or Edge on a computer).
- Build it for real: the React code export follows your links and animations, and Build prompt gives Lovable, Bolt or v0 a description of every screen, link and animation, with a picture of each screen.
Watch each piece of your AI design being drawn
- Piece by piece: while you watch an AI design in your project, what it adds no longer appears all at once. Its cursor visits each new piece in turn: boxes are drawn from their corner, pictures drop into their frames and text is written from left to right, and the cursor says what it is doing ("Drawing a box", "Writing ..."). Your design itself is never held back; this only changes how it is shown.
- Live in ChatGPT too: the live view beside your ChatGPT chat now shows the AI's cursor on the screen being designed. Each new picture is revealed piece by piece over the last one, and between steps it says it is thinking and for how long, so you can see the work is moving. As soon as the first screen is drawn it asks ChatGPT to open it beside the chat, and it says Done when the AI has finished, or Paused, with how long ago it last changed, when the AI goes quiet. New chats always get the newest live view.
- Drawn live on every screen: when the AI works on several screens at once, the live view beside your ChatGPT chat now draws each screen piece by piece too, instead of switching to it all at once.
- Icons stay with their labels: when your AI builds a tab bar, each tab's icon and label are now one group, and so is any icon set beside its label, a logo or a card it draws as one thing. One click picks the whole item; Alt-click still picks a single piece inside it.
- Boards stay apart: a new board your AI adds is never placed on top of another one, so a drawing or a screen no longer shows parts of its neighbours in its pictures and exports.
- Knows when the AI is done: when an AI stops working in your project without saying it has finished, SVG Lab now hands its artboards back to you about two minutes after its last step, instead of showing it at work for up to 15 minutes.
- Decks open cleanly in Google Slides: in a PowerPoint file your AI exports, headings and lines now keep the breaks you designed when opened in Google Slides, instead of wrapping onto an extra line and running into the text below.
- Post images arrive as a ZIP: asking your AI for a deck's square post images now gives you a ZIP of the images, where before the link failed.
15 Oct 2026
Exported text keeps its spacing
- Spaces in your own font: exported SVGs, PNGs and PDFs now carry the space character of every font you use, so words keep the same spacing as on your canvas instead of taking the spaces from a stand-in font.
- Watch your AI design, live: while ChatGPT, Claude or another AI works in your project, its cursor moves to what it just changed and says what it is doing, new shapes fade in where they land, the view follows the artboard it is working on (switch Follow off to look around), and a small card lists its last steps and how long it has been working. Between steps it says it is thinking.
14 Oct 2026
Imported SVGs look exactly as they do in your browser
- What the file shows is what you get: importing or dropping an SVG file, and drawings your AI brings in, now keep rotated and skewed shapes, rounded corners, arcs and curves, reused symbols, clipping and masks, transparency, gradients, dashed lines and mixed text styles. Before, some of these came in flat, moved or missing.
- Drawings stay intact when moved: a clipped, masked or gradient-filled drawing keeps its look when you copy it or fit it into another artboard.
13 Oct 2026
Every picture is drawn from a reference
- No more pictures made of circles (Codex and ChatGPT Work mode): a person, an avatar, an animal or any other picture can no longer be put together from basic shapes straight on your screens. Your AI is stopped and shown how to copy it from a reference picture on its own drawing artboard first, then place it, so the drawings in your screens are real drawings.
- Code exports are checked too: exporting screens as code now passes the same checks as exporting them as SVG, so an unfinished drawing cannot slip through that way.
- No traced artwork: SVG Lab now refuses outlines that look machine traced, from any tool or script, in ChatGPT and Codex. A trace is hundreds of tangled shapes you cannot edit or animate; your drawings are built part by part instead.
- A drawing for each thing your screens show: in ChatGPT Work mode and Codex, each illustration starts from a reference search, then three new versions made with the app's image generation, and is drawn by hand from the best one. A lesson about watercolours gets a painting, not the same picture reused on every screen.
- New projects start clean: a new project no longer carries symbols, gradients or filters over from the project you had open before.
12 Oct 2026
Better drawings in ChatGPT and Codex
- Drawn from a reference: when ChatGPT or Codex designs with SVG Lab, each illustration is now drawn from a reference picture on its own artboard and checked against it, instead of being put together from basic shapes on the screen.
- A new picture, copied exactly: your AI app first makes three altered versions of the picture it found with its own image generator, without stopping to ask, so nobody's picture is used. Your drawing then copies the best version exactly, instead of a loose version of it.
- References stay private: the picture your AI draws from can no longer be placed into one of your projects or exports.
- No more stuck exports: a drawing that matches its reference closely (75 out of 100 or better) no longer stops your screens from exporting. Your AI is told what is still unfinished, with the drawing's real score.
- No skipped references: in Codex and in ChatGPT's Work mode, once your AI loads a reference picture, your screens only export after a drawing copies it. If your AI skips that step, the export tells it exactly what to draw first, so you get the drawing you asked for rather than a rough stand-in.
- Chat mode always finishes: in ChatGPT's Chat mode, an unfinished drawing no longer stops your screens from being placed or exported. You get every screen, and your AI is told which drawing still differs from its reference.
SVG Lab inside ChatGPT
- The editor in ChatGPT's sidebar: with SVG Lab connected to ChatGPT, SVG Lab now appears in ChatGPT's sidebar (the Explore menu). It opens the SVG Lab editor full screen inside ChatGPT, already signed in to your SVG Lab account, with My Projects. Free accounts get it too.
- Live design beside the conversation: in a conversation, open More, then Plugins, then Live design, and the design SVG Lab is making stays open in a panel beside the chat, following each screen as it is made.
- Open in SVG Lab opens your project: ChatGPT's own Open in SVG Lab button on SVG Lab's cards now opens your project in the full editor, instead of the SVG Lab home page.
- Room for the message box: the editor in ChatGPT's sidebar now keeps clear of the space ChatGPT reports as covered, so its zoom bar and status line stay in view, and it shows "Opening SVG Lab..." until the editor is ready instead of an empty dark panel.
- SVG Lab's own icon: SVG Lab's entries in ChatGPT's menus now show the SVG Lab mark instead of a blank square.
Your project opens sooner
- Faster start after sign-in: SVG Lab now opens your saved project while it checks your account, instead of waiting for that check first, so your work appears about a second and a half sooner each time you open the editor signed in, in the browser and in ChatGPT's sidebar alike.
11 Oct 2026
Credits for everything, Starter and Ultra plans, and a plans page
- Credits: SVG Lab now counts in credits, with one balance for the SVG Lab MCP, AI Generate and Auto-trace. A tool call your AI makes costs 1 credit, a reference image search 2, an auto-trace 5 and an AI generation 500. Anything that fails costs nothing. The Billing page lists what each action costs and how many credits you have left.
- New plans: Starter ($7 a month, 2,500 credits, up to 600 app screens) and Ultra ($99 a month, 37,400 credits, up to 9,200 app screens) join Plus ($20 a month, now 7,400 credits, up to 1,800 app screens) and Max ($40 a month, 15,000 credits, up to 3,700 app screens). Every paid plan includes the SVG Lab MCP, and plan credits renew each billing month.
- Top-up packs: $10 for 5,000 credits (up to 1,200 app screens), $25 for 10,000 credits (up to 2,400 app screens) or $50 for 25,000 credits (up to 6,100 app screens), on any plan, Free included. Top-up credits never expire.
- AI Generate with credits: every account's first AI generation is free again. After that, any signed-in account with enough credits can use AI Generate, Free included with top-up credits. Pro and Studio keep their monthly generations as they are.
- 100 free credits (up to 24 app screens): every account still gets 100 free credits (up to 24 app screens) once. Each action is paid from your plan's credits first, then your free credits, then top-up credits.
- A plans page: svglab.app/plans shows what each plan includes, what each action costs in credits, and how top-ups work.
- The same tools on every account (Codex and ChatGPT): an AI app connected at svglab.app/mcp/openai now sees all of SVG Lab's tools even when your account has no credits left. Each tool then tells your AI plainly that it is not available on your account and points to the plans page; nothing runs and nothing is charged.
- Safer reference search: the pictures your AI looks at before it draws are now searched with SafeSearch on, and photos for your design (Unsplash and Pexels) are kept separate from drawing references.
Watch your design being made in ChatGPT and Codex
- A live card in the chat: when ChatGPT starts or opens a project with SVG Lab, a card appears in the conversation and follows the work: the screen being drawn, what it is doing right now (for example "Drawing: cow head"), and every screen so far. Tap a screen to look at it, or keep the card in view while you scroll. It needs no sign-in.
- The real editor beside the chat in Codex: Codex can now open your project in its browser pane as it starts, so you watch every change in SVG Lab itself (sign in there once).
- Finished screens in the chat: when the work is done, your AI exports each screen, and each one shows in the conversation with a button that opens it in SVG Lab.
- Photos or drawings, your choice: your AI now asks once whether your screens should use photos, drawn illustrations or a mix. Drawings start from a reference picture, and a later "use illustrations instead" is redrawn in SVG Lab rather than made in an image generator.
10 Oct 2026
Privacy update: the Billing page
- What changed: on the Billing page only, we now record how the page is used and the steps of a checkout (opened, payment started, paid or closed), so we can see why some people stop before paying and fix it.
- What stays private: everything you type is masked, your name and email are masked, and Paddle's payment form is never visible to us. The editor and your drawings are never recorded. Recordings are deleted after 30 days.
- Details: the Privacy Policy describes exactly what is collected.
Phone screens that fit every phone
- Try your screen on other phones: the artboard size menu now lists phones and foldables (from a small iPhone to a Galaxy Fold, closed or open). Turn on Responsive and pick one, or drag the artboard's edge: rows and cards stretch, the tab bar stays at the bottom, and a list adds a column on a wide foldable.
- Your design is never lost: going back to the size you designed at puts every part exactly where you left it.
Codex can turn your screens into code
- Code you can build on: when Codex designs with SVG Lab (or any AI app connected at svglab.app/mcp/openai), it can now export your screens as HTML, React, SwiftUI, Jetpack Compose, Flutter, or a layered SVG for Figma, with your colours and fonts as design tokens and every drawing and icon as its own file. You get one download link with everything in it.
- A React app that runs as downloaded: run npm install, then npm run dev, and your screens open as a working app. Buttons, tabs and back arrows already move between your screens, and the README lists which tap goes where. A back arrow works even when your AI left it unnamed, and your AI now tells you which taps it wired.
- Build the app in your own project: ask Codex to build your app from a design you made, in the same chat or a later one (give it the project's link or name). It reads how each screen is laid out (rows, columns, spacing, what stretches, the tab bar, where each button goes) and takes your drawings and icons as files, as SVG or PNG, so your code gets your exact design rather than a redrawn copy.
- One file per drawing: an icon or drawing you used in several places (a heart on every card) is now one file in the code export, and every screen uses that file. A copy in another colour or size keeps a file of its own.
- One design across phone sizes: your AI can also make copies of a screen for other phones and foldables, laid out for each one.
- Plans asked up front: before it starts, your AI asks a few plain questions (who the product is for, which platforms, photos or drawings) so the design fits what you are building.
9 Oct 2026
Cleaner drawings when Codex designs with SVG Lab
- Drawings start from a real picture: when Codex draws the illustrations for your screens, it now finds a reference picture to study, copies it closely on a board of its own, and only then places the drawing in your design. The reference stays private: it is never placed in your design or exported.
- Each copy is checked before it goes in: the Design Engine measures the drawing against its reference, faces, hands and small parts included, so flowers, flags and grass come out as real shapes instead of soft blobs. A drawing that is not finished cannot be placed or exported yet.
- Clean, thin lines: hand drawn lines are smoothed into clean curves, a line drawn twice on top of itself is caught, and line weights now shrink with a drawing when it is fitted into a small card.
- Nothing to change on your side: Codex connected to svglab.app/mcp gets this automatically. Other AI apps keep the Design Engine they have today.
7 Oct 2026
Saving fix
- Work in a background tab no longer waits for you to come back: if one slow moment made the editor think it was offline while its tab was in the background, it kept your latest changes unsaved until you looked at the tab again. It now checks the connection on its own and saves.
- Your AI app can start or open a project after a hiccup: when the editor your AI app drives briefly lost its connection, starting a new project or opening another one could keep failing for minutes. It now reconnects, saves your work first, and carries on.
6 Oct 2026
Connect your AI app the moment early access switches on
- Steps for your app, right where you are: once early access is on, the early access page now shows how to connect the AI apps you picked (Claude, Claude Code, ChatGPT, Codex, Cursor, OpenCode or another app), with the others one tap away under Other apps. Claude gets an Add to Claude button that opens Claude with SVG Lab already filled in; Claude Code and Codex get the command or address with a Copy button.
- The steps stay on screen: every link on the panel, including Open SVG Lab, now opens in a new tab, so you can follow the steps while you set things up.
Sign out everywhere
- One button signs out every device: Sign out is now Sign out everywhere, in the editor's account menu, on your Account page and on Billing. It asks first, then signs your account out on every device and browser where you are signed in.
- Disconnect your AI apps at the same time: if you have AI apps connected to the SVG Lab MCP, the same confirmation offers to disconnect them too (ticked by default). If an app cannot be disconnected, SVG Lab tells you which one, still signs you out, and you can remove it from the SVG Lab MCP panel when you sign back in.
SVG Lab MCP fix
- look no longer fails when a shape sits over text: when your AI app checked a screen where a shape overlapped some text, look could stop with "Could not measure the artboard" instead of reporting the result. It now measures the overlap and reports it.
1 Oct 2026
A faster Design Engine for your AI, a support page, and plainer account messages
- A faster SVG Lab MCP: your AI now drives SVG Lab's Design Engine by writing one short script that can build a whole screen in a single step, then looks at the render with measured problems (text running out of its box, overlaps, low contrast, anything past the frame, missing fonts). It decides the look itself, reads the drawing and design guides when it wants them, and gets there in far fewer calls than before. If your AI app still lists the old tools, restart it or reconnect SVG Lab.
- Support page: svglab.app/support lists every way to reach us (email support@svglab.app, or the Send feedback button in the editor) and what to include so we can help quickly. It is linked from the footer of every page and from the editor's status bar.
- Plainer account messages: when your account has no SVG Lab MCP requests left, or reaches a daily or monthly limit, your AI now tells you what ran out and when it resets, without a sales pitch or a billing link. Your plan and requests are on your Account page as before.
- Less account detail in your AI's replies: the connection check no longer repeats your email address or internal account details back to the AI app; it reports your plan and whether the editor is open.
- The privacy policy's contact section now points to support@svglab.app.
30 Sep 2026
Richer designs from your AI: skies that glow, photos where they belong, every font open
- Light and atmosphere are back: your AI can now give a design the depth its subject has: a night sky that fades from deep blue, stars and a moon with a soft glow, a gentle pool of light behind the main picture on a welcome screen, fine grain on a poster. The check still flags the tired look (purple into blue washes, glowing buttons, gradient headlines) but no longer flattens everything else.
- Photos and drawings in the right places: your AI now decides whether your product is led by photos (food, travel, shops), by drawings (kids learning, meditation, budgeting) or by both, and then picks a photo or a drawing for each spot on each screen. A screen can mix them, a photo at the top and drawn icons and empty states below, instead of one screen of stock photos and the next all drawings.
- Room for drawings: list rows can hold a small drawn picture instead of an icon, cards can have a full width picture area at the top, and a picture area with no size given is now generously tall.
- Popular fonts unlocked: the popular fonts every AI reaches for (Inter, Fraunces, DM Sans and others) are no longer off limits. They are listed last, and your AI has to say why one fits your product before it uses it for headlines.
- Finishing no longer stalls on style notes: your AI is held back only by things you would see as broken: text that overlaps or is covered, something outside the frame, a part that came off a drawing, or text too faint to read. Style notes, such as a drawing without detail lines, are still reported so it can fix them or tell you, so it spends its time on your design instead of clearing notes, and it no longer skips drawing a person to avoid one.
- Cards inside cards: one inset panel inside a card is no longer flagged; only a card stuffed with several matching cards is.
- Clearer guidance for your AI: its working instructions are about a third shorter, with new guides on mixing photos and drawings, on what a premium screen looks like in each kind of product, and on when gradients, glow, texture and glass are right. Its replies no longer show a running timer, so it stops cutting corners to save time.
29 Sep 2026
Your AI looks at references, dresses its people and finishes properly
- References without taking over your screen: your AI can now hand SVG Lab a link to a reference picture and see it right in its reply, so it no longer opens stock image pages in the browser tab you are watching SVG Lab in. Before it draws a person it looks at at least one reference, and the project remembers that it did.
- People wear clothes: a person your AI draws is now flagged if nothing covers the torso or hips, and it cannot finish the screen until the figure is dressed for what your product is about (or it tells you why bare skin is right, for a swimmer).
- Finished means checked: when your AI asks for the project link while it still holds your artboards, SVG Lab tells it to run the quality check and hand the artboards back first, so you get the checked version.
- Fewer false alarms: scenery such as rocks in the sea on a map is no longer reported as a part that fell off a drawing, and the detail line check now says exactly what counts.
- Fewer failed steps: when your AI uses a wrong option name, SVG Lab now names the right one ("use patches"), so small fixes such as centering a line of text go through instead of being dropped. The document summary also comes back in pages instead of one very large reply.
- A new colour check: a hot coral or sunset orange accent is now flagged as the colour models pick by habit, so your product gets colours from its own subject. If your AI repaints a screen but leaves the old accent behind, it is told to update it, so a rebuilt layout does not bring the old colour back.
- Every product its own look: SVG Lab now remembers the colours and fonts of your recent projects and tells your AI when a new one starts to look like two or more of them.
- A reference before every drawing: your AI now looks at a reference picture before it draws anything by hand, not only people: animals, objects, plants, clouds and waves too. Until it has studied one in your project, its drawing tools ask it to find one first. Plain lines, dividers, shapes, icons, text and layouts are never held up.
- Drawn only projects: if you ask for no photos, your AI can set the project to drawn only. It then draws every picture itself instead of filling screens with stock photos, still studies photos as references, and the finish check counts its drawings as the pictures a screen needs. The setting is saved with the project.
Outlined text exports whole, and groups keep their names
- Outlined text is complete again: exporting with text turned into outlines, or converting text to outlines in the editor, could cut a line short after a few letters ("Plate 12" came out as "Plat") in some fonts, including Geist, Manrope and Onest. Every letter now comes through.
- Smoother area charts: the shaded area under a chart line now follows the same curve as the line, so there are no pale slivers near the peaks. The label pill on a large chart grows with the chart instead of staying tiny.
- Groups survive export with their names: an exported SVG keeps each group as one group, named after what you or your AI called it, so it opens in other design tools already organised. Your AI can name a group when it makes it, or rename one later.
- References stay put: when two AI sessions work on your account at once, one loading reference pictures no longer pushes out the other's.
- Fewer false alarms from the finish check: leaf stems and veins, strokes inside a drawing your AI grouped as one object, lens rings and small sprouts on a branch are no longer flagged, so your AI stops working around the check instead of drawing what you asked for.
28 Sep 2026
Fewer long waits when a service is slow
- AI Generate now runs on Arrow 2, the newest image and text to SVG model from QuiverAI, the service behind it.
- AI Generate no longer makes you wait through a second attempt when the image service times out. You get a clear "took too long" message straight away, and the try is not counted against your free generation or your plan, so you can try again (a shorter prompt or a simpler image usually finishes faster).
- Auto-trace now gives up on a stuck trace server after two minutes instead of waiting indefinitely, and tells you the trace did not count against your limit.
- SVG Lab MCP: photo search no longer stalls your AI when a photo service is slow, and images your AI hands over in slightly unusual formats (wrapped in quotes, or URL-safe encoding) are now accepted instead of refused.
- When SVG Lab's server editor reaches the end of a long session while it is still saving your work, it now waits for the save to finish before closing.
- SVG Lab MCP: your AI no longer hits timeouts while SVG Lab's server editor opens or when it sends several calls at once. Every call now answers within 45 seconds; if the editor is still opening, your AI is told to send the same call again in a few seconds. Calls sent together now take turns instead of running over each other, and reading design guides now counts as activity, so the server editor does not close in the middle of a task. Your AI can also pause the server editor itself before it stops to think, and it reopens on the next drawing step, so you do not use up design time while nothing is being drawn.
The Design Engine checks the finish before your AI hands it over
- Drawings that read: animals, people and objects your AI draws now carry a few lighter detail lines (an ear, a leg, a rim) so you can tell what they are, and handles and ears stay attached to the body.
- Photos where they belong: the SVG Lab MCP now expects real photographs for the things people browse, buy, visit or meet, alongside drawings, icons and text, instead of drawing everything.
- No filler: loose arcs, rings and swooshes that do nothing are flagged, and so is text hidden under a badge.
- More variety: fonts and colours now come from your product instead of the same cream and red look on every project.
- Status colours match your palette: the green, amber and red used for success, warning and danger badges, alerts and charts are now derived from the accent your AI chose for the product (same hue family, your palette's saturation and warmth) instead of one fixed set shared by every project. Set them yourself in design_tokens to override. The MCP's session_status now reports the daily design-time cap that actually applies to your plan.
- Your AI cannot export or finish a screen while one of these is open, unless it tells you why the design needs it.
27 Sep 2026
Try the SVG Lab MCP free: 100 requests on every account
- Every SVG Lab account, Free included, now gets 100 free requests to try the SVG Lab MCP. There is no form to fill in: add SVG Lab to your AI app, sign in, press Approve, and it starts designing. The setup guide has the steps for every app.
- Ask for SVG Lab MCP access on the early access page and it is on straight away, until 2 October. An email with the setup steps follows within a minute.
- New top-ups: buy extra MCP requests on the Billing page on any plan. Top-up requests never expire and nothing renews.
- The free requests are a one-time trial per account; they do not renew each month.
- Updated billing policy: all purchases, plans and top-ups, are final and non-refundable. You can still cancel a subscription any time and keep access until the end of the period you paid for.
- Design time on SVG Lab's servers is now counted per account, never shared: how much you get each day depends on your plan, and nobody else's use can use up yours.
- Each tool call your AI app makes is one request, and calls that fail are not counted. Plus and Max use their monthly allowance first, then the free requests, then top-ups. Your assistant can tell you how many are left, and the Billing page shows your free requests and top-up balance.
See your MCP usage, live
- The Usage button in the editor now shows your SVG Lab MCP use too: your plan or early access allowance used this period, free requests left, your top-up balance, and your top tools and projects.
- Press More for the new MCP usage page: requests by allowance, by tool, by project and by day, a list of recent calls that updates every few seconds, and your top-ups with where those requests went.
- Only counts are kept: the time, the tool, which allowance paid, the project and whether the call worked. Nothing your AI app builds or sends is recorded, and the history is kept for 90 days.
26 Sep 2026
New plans: the SVG Lab MCP comes with Plus and Max
- The SVG Lab MCP is now open to everyone on two new plans. Plus ($20 a month) includes 5,000 MCP requests a month and Max ($40 a month) includes 15,000. One tool call your AI app makes is one request, and calls that fail are not counted. Your assistant can tell you how many requests are left and when the allowance resets.
- Plus and Max also carry the full editor: unlimited exports, 250 cloud projects, up to 50 pages per document and 250 auto-traces a week, with 25 AI generations a month on Plus and 80 on Max.
- Free and Pro ($3.99 a month) stay as they are, for designing in the editor. Studio is no longer offered. See the plans.
- Early access accounts keep their access, and early access requests stay open until Friday 2 October 2026. MCP sessions are no longer reviewed; see the privacy policy.
No more pairing codes
- Once you connect your AI app, it designs in SVG Lab straight away. You do not need SVG Lab open, and there is no code to type in: your assistant works in its own SVG Lab editor on our servers and saves everything to My Projects, where you can open, edit and export it. The Connect to AI button is gone from the top bar.
- The SVG Lab MCP item in your account menu now opens one small panel: whether early access is on for your account (with a link to request it if not), your connected apps with a Disconnect button for each, and the setup guide. A key for command line apps is still there under Use a key instead of signing in.
Form lines for illustrations drawn through the MCP
- Your AI assistant can now give a flat, one colour object such as a vase, jug, mug or bottle the few curved lines an illustrator would draw: the ellipse of the rim, the turn of the neck, a highlight down the widest side and a ring at the foot. SVG Lab fits each line to the object's real outline, keeps it inside the shape, and groups it with the object, so one undo removes them all.
- When your assistant reviews an illustration, SVG Lab now points out a part that floats clear of the object it belongs to, such as a mug handle a few pixels off the mug, and says how far to move it. Things that are meant to float, such as steam or sparkles, are left alone.
- Fixed: a new project started by your assistant could carry over the colours and layout notes of the project it was working on before. Every new project now starts clean.
- Fewer refused requests: small differences in how assistants describe a shape (an opacity written in a different place, a colour given by its design name, a mirrored half marked as closed) are now understood, with a note saying how they were read.
- The editor's welcome now starts with the SVG Lab MCP: you don't have to draw it yourself. Connect Claude, Codex or Cursor, ask for app screens, icons and illustrations, and SVG Lab's design engine builds them in your SVG Lab projects, and you watch them appear. A Request early access button takes you to the request page, and Explore the editor carries on with the tour. Signed in, a short pointer shows where to ask for access: SVG Lab MCP: request early access in your account menu.
- The early access page now opens with real screens made through the SVG Lab MCP, right next to the headline, so you can scroll through them straight away.
- Fixed: moving or resizing a shape could stop responding and fill the console with errors if the shape was removed during the drag, for example by an undo or by your AI assistant working in the same project.
See where your AI assistant is working, and pause it
- While an AI assistant works in your project through the SVG Lab MCP, the artboards it is using get a coloured frame and a label with the app's name (Claude, ChatGPT, Codex, OpenCode or Cursor) and what it is doing right now. The artboard list marks them with a small lock.
- Those artboards are the assistant's until it finishes: you can look, zoom, select, copy and export, but changes wait. Everything else in the project stays yours, including new artboards.
- Press Pause or Stop on the label to take the artboards back straight away. The assistant is told and stops working there until you ask it to carry on. If it goes quiet for 15 minutes, the artboards come back to you on their own.
- When you change something while it is paused (move an artboard, add an image, restyle a shape), the assistant is shown exactly what changed when it comes back, tells you in a sentence, keeps your changes, and only puts things back if you ask.
- Fixed: an assistant connected before this update could not carry on after you pressed Pause or Stop, because its app still had the older list of SVG Lab tools. It now picks up again when you ask it to continue, with the tools its app already has, and still tells you what you changed while it was paused. SVG Lab also asks connected apps to refresh their tool list after an update.
- Two assistants can work in one project at the same time, each in its own group of artboards, with its own colour.
- While an assistant designs in a project you have open, your window switches to watching: a line at the top says so, its changes appear as it makes them, and your window leaves its artboards to it instead of saving them too, so nothing gets overwritten and your window stays quick. You can keep working on your own artboards and add new ones. Take over or Pause hands everything back to you at once.
AI apps stay signed in
- Fixed: an AI app connected by signing in (Claude, ChatGPT, Codex, Cursor, OpenCode) could be signed out after a few hours, usually when it ran several tool calls at once or while we were updating SVG Lab, and you had to connect it again by hand. A connection now stays signed in for months while you use it, and a brief interruption on our side makes the app wait a moment and try again instead of signing you out.
- Disconnect still switches an app off straight away.
- New setup guide: step by step setup for the SVG Lab MCP in Claude, Claude Code, ChatGPT, Codex, Cursor, VS Code, OpenCode, Windsurf and Gemini CLI, with copy-paste settings, a prompt that checks it works and fixes for common problems. It is written so your AI assistant can read it too and set SVG Lab up for you.
- Connected apps are signed out automatically if an old sign-in is ever reused, which protects your account if a token leaks.
One project open in several windows
- Two windows on the same project no longer overwrite each other: each artboard saves on its own and updates live in your other windows, so you can keep working on one artboard while another window (or your AI assistant) changes a different one.
- Your work elsewhere stays put when an update arrives: your selection, your undo history and any change you have not saved yet on other artboards are left alone, and undo never takes back a change made in another window.
- If the same artboard was changed in two windows at once, both versions are kept: the one saved first stays, and yours appears beside it as "(your copy)", with a message saying so. Nothing is thrown away.
- Changes your AI assistant makes are saved about a second after each step, so you can watch them arrive in your own window.
- Faster editing on big projects: scrolling, panning, zooming and selecting no longer start a save, and a save now only prepares the artboards you actually changed. Your own edits save about two seconds after you stop, never in the middle of a drag.
A lighter editor while you think
- The editor now does almost nothing while you are not drawing. On a project with 500 shapes, the glowing buttons and tabs used to keep the page busy about a quarter of the time even when nothing changed; they now look the same and cost close to nothing, so big projects feel smoother and laptops stay cooler.
25 Sep 2026
Connect an AI app by signing in
- Connecting Claude, ChatGPT or Codex to SVG Lab no longer needs a key pasted into a web address. Add https://svglab.app/mcp as a connector, and your AI app opens a Sign in to SVG Lab page. Sign in, check which app is asking, press Approve, and you are connected. Command line apps such as Codex CLI and OpenCode can still use a private key.
- The MCP window in the editor now lists your Connected apps, each with a Disconnect button that switches that app off straight away.
- In AI apps that can show interactive results inside the chat (MCP Apps, supported by apps such as Claude, ChatGPT, VS Code and Cursor), a design your assistant exports now appears as a small card: the design itself, the project and artboard name and size, and an Open in SVG Lab button that takes you straight to the project. Where the app allows downloads, the card also has a Download SVG button. Other apps show the export exactly as before.
- When your early access is switched on, you get an email that says so and tells you how to connect.
- The privacy policy and terms have a new section on the SVG Lab MCP: what it receives, how you connect and disconnect, and, during early access, that we review MCP sessions to improve it.
- Fixed: AI apps connected by signing in could be signed out after about an hour and start answering "Unauthorized". It happened when the app renewed its connection from two requests at once, which apps such as OpenCode, Cursor and Claude do when they run several steps in parallel. Renewals are now handled one at a time and every parallel request gets the same result, so the connection keeps working. If an app was signed out before this fix, reconnect it once.
24 Sep 2026
63 new fonts, and outlines for Arial, Georgia, Times and Courier
- 63 new fonts in the Text panel's font list, grouped into Sans, Serif, Mono and Display: expressive display faces for posters and headlines, sturdy text faces for interfaces, and a few scripts for a single word. Every one is free to use in anything you make. A font downloads the first time you pick it, so the editor opens as fast as before.
- Projects that use one of the new fonts load it again when you reopen them, and Convert to outlines works for all of them.
- Fixed: PNG exports of text in a web font (Inter, Lobster, Playfair Display and the rest of the list) came out in your browser's stand-in font instead of the one you picked. Exports now carry the fonts they use, so PNGs match the canvas and downloaded SVG files show the right lettering on any computer, even one without the font installed.
- Convert to outlines now works on text set in Arial, Helvetica, Georgia, Times New Roman and Courier New. Those fonts are not hosted online, so SVG Lab outlines them with an open font that has exactly the same letter widths (Arimo, Gelasio, Tinos, Cousine) and the status bar tells you which one it used. The lettering stays where it was; only small details of the letterforms differ.
- Fixed: converting text in a font SVG Lab cannot outline (Verdana, System UI) left the status bar stuck on "Converting to outlines…" with no explanation. It now names the font and says what to pick instead.
- Fixed: bold or italic on a font that only comes in one weight (Lobster, Pacifico, Bebas Neue and the other display faces) made outlining fail. It now converts using the font's own weight.
Ask for early access to the SVG Lab MCP
- There is a new MCP early access page. It shows what the MCP does with a live demo and real screens a model designed through it: Claude, ChatGPT, Codex, Cursor, OpenCode and other AI assistants design and draw real, editable SVG in your SVG Lab account, with no tab open, and everything they make is saved to My Projects.
- Requesting access is one short step: pick what you want to make (app screens, icons, illustrations, social graphics, diagrams, slides), say which AI app you use if you like, and add a note. You need to be signed in, and the page lets you sign in with Google or an email link without losing your answers.
- You get a confirmation email straight away, and another one when your access is approved and switched on.
- In the editor, the MCP item in the account menu now shows Early access if your account is not in the release yet, and takes you to that page instead of a window with greyed out buttons. Once you have asked, it says Requested. Accounts that already have the MCP see the same MCP window as before.
- There is also a new MCP server page that explains in plain words what the SVG Lab MCP server does, which AI apps it works with, what it can make, and how to get access. You can reach it from the editor's status bar and from the About page.
23 Sep 2026
Connect an AI app to SVG Lab
- Update, 25 Sep: Claude, ChatGPT and Codex now connect by signing in. Add https://svglab.app/mcp as a connector and approve; no key needed. The key steps below still work for command line apps.
- There is a new MCP item in the account menu. It opens "Connect an AI to SVG Lab", where you can make a private key and copy ready-made setup for Claude (Settings, Connectors, Add custom connector), Codex and OpenCode. After that, click Connect to AI in the top bar and give your assistant the code it shows: what it draws lands on your canvas, and you can edit or undo all of it.
- One key per account, shown once. Regenerate replaces it and Revoke switches it off, within a minute, for every app that uses it.
- We are opening this to a small group first. If your account is not in it yet, the MCP window says "Production release pending" and the key buttons stay greyed out.
- For the AI on the other end, SVG Lab now builds arms, legs and hands (open, fist, or gripping an object) from a pose description, checks that parts actually touch, and returns exact geometry, so a plain sentence like "a bent arm holding a wrench" comes out legible instead of as a mitten on a bar.
- Later the same day, from the first real session: an AI can now say "hold this shape" and SVG Lab works out where the fingers wrap, warns when an open hand is lying over a tool instead of holding it, draws a shoulder that grows out of the body instead of a ball on top of it, and gives arms and hands the same outline as the body they attach to.
20 Sep 2026
Big projects save properly
- Projects are now compressed before they are saved, so a big drawing takes a fraction of the space and a fraction of the time it used to. A 2.3 MB project measured here now saves as 0.5 MB.
- Photos you drop into a drawing are stored separately from the drawing itself, privately to your account, instead of being packed into the project. Nothing changes in the editor or in your exports, the project just stops being enormous.
- A slow save is no longer reported as a failed one. If a save is taking a long time, SVG Lab now checks whether it actually went through before saying anything, and retries on its own if it did not. You only see a failure when the save really failed, and the message then tells you what happened, how big the project is, and that your work is still open in the tab.
- The topbar shows how much space the current project takes in the cloud. Past 3 MB it turns amber with the two things that help (delete artboards you no longer need, or split the project in two), and past 8 MB a save is declined with an explanation rather than attempted and failed.
- Fixed: the save chip could stay on “Saving” after a save had already finished.
- Saving no longer costs the size of the whole project. SVG Lab now stores each artboard separately and sends only the artboards you actually changed. On a 2.3 MB project measured here, changing one piece of text used to upload the entire drawing and now uploads 50 KB, which is 81% less. Nothing changes in how you work: projects you already have convert themselves the first time you save them.
- The save chip tells you more while it works: “Saving 2 of 11 artboards” while it uploads, then “Saved just now”. If something does not reach the server it says exactly what is outstanding, for example “Not saved: 1 artboard pending”, gives you a Retry button, and reminds you the work is still open in your tab. It never says a save failed when the save may in fact have gone through.
- Saving runs in the background and never blocks drawing, and a burst of quick edits is grouped into one upload instead of one per change.
- Renaming a project now writes just the name instead of re-uploading the whole drawing, so it is instant on projects of any size.
- If part of a project does not come back from your account, SVG Lab stops saving that project instead of saving over the part it could not read, tells you which artboard is affected, and offers to go and look again. Switching to another project in the middle of a save can no longer mix the two.
- Very large pieces of artwork, a traced image or text turned into outlines, are now kept separately from the project, privately to your account, and fetched before the drawing appears. A project stops growing every time you trace something, and the artboard holding it stops re-uploading it whenever anything else on that artboard moves.
- The project size in the topbar is now a button. Press it to see where the space actually goes: every artboard by name with its size, largest first, plus a row for the shapes that sit outside every artboard, which is often the biggest thing in a project and the easiest to clear out.
- If a tab closes or crashes while a save is still on its way, SVG Lab now keeps a copy of that save in your browser and offers it back the next time you open the project: Restore unsaved changes, or Discard. The copy is deleted the moment your account confirms the save, it is never kept while you are signed out, and it is removed when you sign out.
- SVG Lab now loads inside directory previews such as BuildHop, which used to show a blank page. If you open the editor inside another site, a slim bar at the top offers a link to the full editor in its own tab.
- The editor appears far sooner on a first visit. It used to wait for two files served by other companies before it drew anything on screen, and on a slow connection that was several seconds of a white page. Neither file holds anything up now, so the editor paints in about a second on a mid range phone on mobile data, measured here, instead of five to nine seconds.
- SVG Lab no longer needs any outside service to be reachable in order to open. The sign in library is served by SVG Lab itself, and error reporting is loaded off to the side. If a company firewall, a public wifi login page or a country level filter quietly blocks one of them, the editor still opens and still works. Before this change that gave you a blank page with no explanation, which looked exactly like the site being down.
- Large projects open and respond much faster. A project of several thousand shapes used to spend most of its opening time searching the canvas over and over, and clicking a single shape could take seconds. Opening a 4,890 shape project measured here went from 8.4 seconds to 1.5 seconds, and clicking one shape in it went from 5.0 seconds to 0.9 seconds.
- Undo stops eating memory. Every edit used to keep a full copy of the whole drawing, and SVG Lab kept two hundred of them, so on a 1,630 shape project fifty edits held about 313 MB in the tab and the browser eventually slowed to a crawl. An edit now remembers only the shapes it actually changed. The same fifty edits measured here hold about 15 MB, and undo now remembers up to 500 steps instead of 200.
- Ctrl+Z is fast again on a big project. It used to throw away every shape on the canvas and draw the whole drawing again, which took 3.7 seconds on a 1,630 shape project and about half a minute on a 4,890 shape one. It now puts back only what the step changed: half a second and a tenth of a second respectively, measured here.
- Dragging text is smooth. Moving a shape cannot change its size, but SVG Lab was re-measuring every piece of text you moved on every single frame of the drag. Dragging 100 pieces of text on a 1,630 shape project went from 694 ms per frame to 23 ms per frame, measured here, and dragging 100 shapes of any kind went from 248 ms to 37 ms. The real measurement still happens once, when you let go.
- Artboard changes undo properly now. Resizing an empty artboard, renaming it, changing its background or adding and removing one used to leave Ctrl+Z with nothing to take back, so it skipped past and undid your previous shape edit instead. Each of those is its own step.
- Changed: switching pages now starts the undo history fresh. Before this, Ctrl+Z after a page switch could drop the other page’s shapes onto the page you were looking at.
- Selecting a shape is instant on a big drawing. Every click used to throw away the whole layers list and build it again, row by row, even though nothing had changed but which row was highlighted. The list now only moves the highlight. Clicking one shape measured here went from about 100 ms to about 2 ms on a 1,630 shape project, and from 1.7 seconds to 5 ms on a 4,890 shape one.
- Bursts of changes are drawn once. A change that touches many shapes at once, an alignment, a paste, a colour applied to a whole selection, used to redraw the layers list once per shape. Twenty changes in a row now cost one redraw instead of twenty.
- Dragging is smooth on a big project. Dragging 100 shapes around a 1,630 shape drawing measured here went from 13 frames a second to 54, and dragging 100 pieces of text went from 5 frames a second to 60. Opening the same drawing went from 430 ms to 390 ms, and Ctrl+Z from 770 ms to 110 ms.
18 Sep 2026
Faster opening for big accounts
- Opening SVG Lab while signed in is much faster when your account holds a lot of projects. The editor now downloads only the drawing it opens instead of every project you own, so the “Cloud sync is still taking too long” message is gone.
- Account limits load faster too: checking your plan usage no longer downloads your newest drawing behind the scenes, which took a couple of seconds for big projects.
- Fixed: if a project was deleted on another device at the exact moment you opened SVG Lab, the editor could show “Cloud project did not load” with an empty canvas. It now opens your next project instead.
- Safer saves across devices: a save now only goes through if nobody else changed that project since this device last loaded it. If another device saved first, SVG Lab shows you that newer version with a note instead of quietly writing over it. This closes a case where a slow connection could overwrite work you had just saved elsewhere.
- Live updates across devices: rename, save or delete a project on one device and every other signed-in tab sees it within a second. A rename updates the title in place without touching your canvas; a drawing saved elsewhere replaces the stale copy; the My Projects list refreshes on its own while it is open.
25 Aug 2026
AI Generate is live
- AI Generate: describe what you want, or upload an image, and get an editable SVG placed straight onto your artboard. It sits in the editor topbar. Available on paid plans; free users get a preview and a one-time $3 export unlock.
- Studio Annual now grants its full year of 480 AI generations upfront instead of drip-feeding them monthly.
- Fixed a quota bug that was silently refusing metered requests, and made failed generations give your quota back when the provider did not charge.
24 Aug 2026
Export fidelity overhaul
- Gradients, patterns, filters, clip paths and masks now export as they are, instead of being flattened.
- Text on a path exports as a real
textPath element, so it stays editable and bound to its curve in other tools.
17 Aug 2026
The blog grows to seven posts
- New guides on the blog: resizing SVGs, SVG to JPG, SVG to PDF, SVG to PNG quality, SVG path editing, and changing SVG colors, alongside the founder story that launched it.
- Every post now has its own social card image.
6 Aug 2026
Five free tool pages + the blog launches
1 Aug 2026
Per-artboard locking
- Lock any artboard to protect its contents while you work on the others.
21 Jul 2026
Calligraphy strokes, Fill Region, Magnet snapping
- Tapered stroke ends with independent controls per end, calligraphy style.
- Fill Region bucket tool, Live Paint style: click any enclosed region to fill it, even across overlapping paths.
- Magnet v2: path endpoints snap to nearby strokes, with dual-endpoint rigid snap and an explicit Snap command.
12 Jul 2026
Path editing and billing polish
- Direct select now works correctly on rotated paths, and rotation is baked into exported geometry for every shape.
- Bigger anchor hit targets and tighter path selection hit testing.
- Layers panel: per-row delete button.
- Change plans in place with proration, no cancel-and-resubscribe.
26 Jun 2026
Auto-trace tuning
- Retuned auto-trace presets: stable results, better color fidelity, and more detail preserved at the maximum setting.
- Traced assets now persist reliably to cloud save, and the trace review flow is smoother.
19 Jun 2026
Accounts and support
- Support requests from the in-app widget route to email, with branded reply templates.
- Set a display name in your account settings.
21 May 2026
About + Changelog pages, custom 404
- New About page: what SVG Lab is, who's behind it.
- This Changelog page.
- A branded 404 page so broken links don't dump you on Cloudflare's default.
20 May 2026
True mesh gradient + PostHog analytics
- Mesh gradient editor. Real per-node bilinear color blending, not the radial approximation we shipped first. Grid sizes from 2×2 up to 5×5, with color resampling when you change the grid size.
- PostHog product analytics wired in: cookieless, anonymous, no session recording. See the Privacy Policy for the full disclosure.
19 May 2026
Five new shape tools + compound-path overhaul
- New shape tools: Triangle, Polygon, Star, Rounded Rectangle, Arrow. All with adjustable corner rounding via a slider in the properties panel.
- Text on Path rewrite: drag text along a path to slide it, toggle Above / Below, optional upside-down flip.
- Compound paths first-class everywhere: Knife, Smooth, Offset, Add/Delete Anchor, and Corner↔Smooth toggle all respect holes now (a donut with a hole stays a donut when you cut, smooth, or offset it).
- Symbol editing: detach an instance to edit it. When you re-make the symbol, every other instance in the document updates automatically.
- Code-panel import fidelity: pasting SVG into the live code panel now preserves gradients, patterns, filters, clip-paths, masks, symbols, and text-on-path bindings.
- Defs cleanup: deleting a shape now sweeps any orphaned gradients, patterns, filters, clips, or masks it referenced.
17 May 2026
Milestone A: friends launch ready
- Sentry error monitoring: every JavaScript error and crash is now logged so we can find and fix things you'd never bother to report.
- In-app feedback widget (bottom-right of the editor): type, hit submit, no email needed.
- Mobile touch support via Pointer Events.
- 4-panel onboarding modal for first-time visitors. Skippable. One-time.
- Examples menu with three preloaded designs so a fresh visitor isn't staring at an empty canvas.
- SVG favicon + Open Graph + Twitter Card meta tags: links to svglab.app now have a proper social preview.
- Cloudflare Web Analytics enabled (cookieless, privacy-respecting).
- Knife tool now respects shape rotation, so tilted shapes cut at the right angle.
- Chained boolean operations now preserve holes from earlier operations.
16 May 2026
SVG Lab goes live
- Live at svglab.app on Cloudflare Workers + Static Assets.
- Rebrand: "SVG Studio" → "SVG Lab" everywhere user-visible.
- Initial browser auto-save shipped here; it has since been removed in favor of tab-only unsigned work plus account cloud save.
- Privacy Policy and Terms of Service pages.
- 21 advanced SVG features: linear / radial / true mesh gradients, patterns, SVG filters, clip & mask, symbols, multi-page artboards, boolean ops, blend modes, align & distribute, eyedropper, path offset, text on path, auto-trace, PNG and PDF export, code panel, rulers, and more.