Grid layouts, mixed text styles, Wonder Capture, and a more capable MCP

Block, inline, and grid layouts on the canvas, different styles inside one text layer, the Wonder Capture browser extension, a more capable MCP, and a long list of improvements and fixes across the app.

Features

• Block, inline, and grid layouts. The canvas now supports native document flow. Switch any frame between freeform, flex, block, and grid from the Display control. Grid comes with column and row track editors, a visual dimension picker, gap handles, cell-aware drag and drop, and resizing that snaps to spans. Frames can live inside a line of text as inline elements and wrap with it. Pasted captures keep these layouts, and exported code uses them instead of faking them with flex.
• Style a few words differently inside one text layer. Select characters while editing and change their weight, size, color, or spacing right in the properties panel, so a bold stretch or a colored word no longer forces you to split a sentence into separate layers. The panel lists the styles applied to specific words, editing the base style still flows through them, and Group into one text merges text layers you had split apart back into a single layer.
• Wonder Capture. A Chrome extension that grabs any webpage, or just part of it, and pastes it onto the canvas as fully editable layers. Fonts, images, icons, masks, and typography come along, including from Google Slides and local dev servers. Install it from the downloads page and read the guide.
• A more capable MCP. Add Wonder to Claude as a custom connector with the new guide. Agents can upload local images, generate images and SVGs, open a file in a desktop tab, and work on any page of an open file. Tool calls target the file you are looking at, results are leaner, errors say whether a retry will help, and viewers can use the read-only tools.
• Admin role. Make a member an Admin to manage people and settings without owner-only powers like renaming or deleting the workspace.
• 100 free shaders. The free tier grows from 24 to 100 presets, premium presets show as locked in the gallery, and two library updates brought new presets.
• New AI models. Gemini 3.7 Flash and Grok 4.6 join the model picker.

Workspace improvements

• Recent files in a new tab. In the desktop app, the + button opens a recent-files view with a New file action and search.
• Open in desktop from the browser. Files hand off straight into the desktop app, and the button only shows when you are in the browser.
• A steadier desktop app. AI activity shows in tabs, tabs are easier to tell apart, external links open in your system browser, an offline screen and a crashed-tab placeholder replace blank views, and the open file and window size survive an auto-update.
• Small additions. Discord and Slack links on the onboarding welcome screen, and new files are named after famous designers instead of cities.
• More docs. New guides for variables and tokens, icons, vectors, shaders, pasting your existing designs, Wonder Capture, and your codebase.

Editor improvements

• Copy a link to your selection. Share a link that opens the file on exactly the elements you selected, and the share button now copies the current page link.
• Paste into everything selected. Paste fills every selected element, and paste-and-replace swaps each one while keeping its anchor.
• Hold Space to keep elements inside their frame. Move things around without reparenting them.
• More shortcuts. Opt+C and Opt+V center an absolutely positioned element in its parent, Shift+A wraps the selection in a flex frame, and every canvas tool and grouping action is searchable from the command palette.
• Multi-selection. Spacing helper lines, shader fills, and anchor changes now work across a multi-selection.
• Color picker. Bind color variables from inside the picker, the fill row has its opacity input back, and the fill-type switcher hides when there is only one fill.
• Icon library. Add selected icons to your library, organize them by dragging, multi-select, and subcategories, and preview icons on hover in the picker.
• Dark canvas in dark mode. New files default to a dark canvas background when the app is in dark mode.
• Layout controls. The properties panel shows and edits content-sized, percentage, and reverse-wrapped layouts, text gets indent and font smoothing controls, and duplicating again repeats the previous offset.
• Faster canvas. Toggling sidebars, entering frames with many layers, marquee selection, and dropping images are all faster. Selection hides while you adjust shadows, and the layers panel has a horizontal scrollbar.
• Presence. The avatar strip is now a multiplayer dropdown, multiplayer tools no longer crowd each other, and a lone collaborator shows their avatar instead of +1.

AI improvements

• Multi-page generation lands on the canvas. Each generated page is placed next to the others instead of on top of them, whether it comes from chat or from MCP.
• Layouts the AI can use. Designs use grid, block, and inline layouts where they fit, not only flex.
• Better design fundamentals. Palettes derived from a mood, a deliberate typographic register, and fewer clichéd pairings.
• Predictive text knows the conversation. Text completion in chat uses the AI's last reply as context.
• Image generation. Prompts pass through verbatim, and reference images are fetched from your uploads.

Bug fixes

• Fixed selection boxes being offset on generated elements, and the selection box not working when a drag starts over the parent.
• Fixed flex fill items not taking equal space, and changing text properties before typing removing the text box.
• Fixed dragging one gradient endpoint moving the other, and mask overlays not matching gradients.
• Fixed duplicated fixed frames losing their fixed position, and pinned children losing their anchor when reparented or alt-drag duplicated.
• Fixed nested frames breaking drag-back reparenting, clicks under stacked frames selecting the wrong frame, and dragging a child out of an overflowing parent reparenting it.
• Fixed two copies of one SVG sharing definitions, SVG masks lost on capture paste, and an SVG icon replaced via chat becoming unselectable.
• Fixed custom fonts not applying after upload, captured designs losing their typography after reload, and custom fonts missing from screenshots and thumbnails.
• Fixed pasting a large captured design failing, Tailwind v4 pages refusing to paste, layered CSS masks lost through capture, and centered text lost on import.
• Fixed shader and SVG images missing from exported code and image exports, the ASCII shader rendering blank in screenshots, and rotated text clipped in local export.
• Fixed chat images applied at oversized sizes, uploaded images dropped when one was slow, and Lucide icon requests silently failing.
• Fixed generated desktop pages rendering squished, and a prompt refused for content policy telling you to contact support.
• Fixed editing an older chat message closing when changing the model, queued messages appearing before the current reply, and truncated answer options.
• Fixed collaborator cursors drifting from their real positions, some users never connecting to realtime collaboration, files that loaded empty or failed to load, and very large files failing to open.
• Desktop: fixed keyboard shortcuts after refocusing, a second login opening the welcome file, expired sign-in links dead-ending, registration not returning to the app, GitHub sign-in doing nothing in settings, missing file thumbnails, the home tab not syncing, and the update prompt reappearing after a restart.
• Fixed toasts rendering behind modals, focus states on properties inputs, the zoom menu and screen picker layouts, the repo list scrollbar overlapping Connect, and text not being copyable from the MCP modal on desktop.
• Fixed new files missing from search, the last created file hidden on the desktop home tab, and the loading animation appearing late when opening a file.
• Fixed sign-in errors rendering nothing, a changed provider email locking you out, and a disconnected sign-in method still signing in.
• Fixed MCP authentication expiring after a computer lock, and tool calls targeting the wrong file with several tabs open.
• Fixed unsupported browsers being blocked before reaching the canvas, and the password field rendering badly in Firefox.