What's New

August 12, 2026

The first post-launch update — a new video model, node references you can write into a prompt, server-side exports, canvas blend modes, and a long round of fixes.

New

• Seedance 2.5 — New video model for text-, image- and reference-to-video, and the only one that renders past 15s (up to 30s).
• Node references as pills in your prompt — Selected board nodes appear as pills inside the prompt text, so you can weave them into a sentence. Uploads get numbered pills, and pill order is the reference order.
• "Use as Prompt" restores image references — The canvas node selection and its drag order come back with the text, so you can iterate a video prompt without re-selecting and re-ordering images.
• Server-side image export — Images render on the server instead of on your GPU, so large and 2x exports no longer depend on your machine or tab memory. Video export loads only the node being exported instead of the whole board.
• Uploads keep their real dimensions — Dropped, pasted, and uploaded assets insert at their intrinsic size instead of being scaled down to a 1024px longest edge.
• Figma blend modes render on the canvas — Multiply, Screen, and Linear Dodge now composite correctly on nodes, and a blend mode set on a group carries down to everything that draws inside it.
• Group and ungroup from the canvas context menu — Right-click a selection for group/ungroup, with sibling selection resolved through the parent for multi-node selections.
• Better SVG rendering — A new vector codec replaces the old parser, so imported SVGs render closer to the original.
• Split an image into a grid — Ask the agent to cut an image into rows and columns; each tile becomes its own editable node, ready for per-tile edits or animation.
• MCP uploads without S3 access — Clients that cannot reach S3 directly get an upload card instead of a failed call.

Fixed

• Boards and long chat histories load faster — thumbnails render only for what you can see, and never in the middle of a drag.
• Chat history scrolls all the way to the last message, and node reference thumbnails fill their tile again.
• Adding a node pill no longer pulls focus away from the board.
• Fonts you upload are usable as soon as they finish processing — caches invalidate when the font is ready, the agent refreshes its font list on a family miss, and a font reference it cannot read no longer breaks the node.
• Background blur no longer corrupts or discolors other layers on the board.
• Copy-paste centers on the true bounds of what you copied, and pasted nodes come back selected.
• Group resize handles are back on timeline preset rows, and deleting a video node removes its time track with it.
• The agent no longer claims an animation it did not make, and it surfaces the animation on the canvas when one does land.
• Export is disabled for multi-frame video selections instead of failing later.
• Ask the agent about an image and it reads the asset's real size and resolution instead of estimating.
• Asking the agent a clarifying question no longer leaves the conversation stuck.
• Reference portraits tell you when registration is not yet active instead of failing quietly.
• Reve 2.1 is off the image model list for now.

August 4, 2026

Kyoso is now publicly live — anyone can sign up and start creating, no invite required. Here's everything that shipped with the public launch.

New

• Kyoso is public — Open self-serve signup is live; create an account and start creating with no invite or waitlist.
• Shareable signup/login links — /signup and /login are direct shareable URLs you can send to anyone.
• Promo codes at checkout — Apply a promotion code when subscribing or buying credits.
• Reve 2.1 — New text-to-image / image-to-image model in the picker.
• Reference portrait generation — Generate a consistent portrait of a person and reuse it across images and videos.
• Per-glyph, word & line text animation — Animate text at a finer grain, targeting individual glyphs, words, or lines.
• Engine-level undo/redo — Cmd/Ctrl+Z, Shift+Z, and Ctrl+Y drive the canvas engine's own history, including text edits and effects, and stay reliable through complex and multiplayer gestures.
• SVG on the board — Drop, paste, or upload (via URL) vector assets that render with full parity.
• Group dragging — Move a group of selected objects together as one.
• Snapping transform handles — Resize and transform handles snap into place; hold shift to constrain.
• Multi-selection styling — Edit typography, fill, and stroke across a multi-selection, with "Mixed" indicators where values differ.
• Multi-selection alignment — Tidy up, distribute spacing, and a new gap field.
• Frame nodes from the layers menu — Group nodes into a frame via the layers menu or Cmd+Option+G.
• Canvas styling upgrades — Blend modes, stroke caps/joins/miter, and radial gradient positioning.
• Richer Brand Hub rendering — brand.md renders doc references as pills, hex colors as swatches, and asset references as thumbnails.
• Center canvas on a referenced node — Click an attached node thumbnail in chat to jump the canvas straight to it.
• Agent can suggest ungrouping — The agent recommends and performs "unflatten" to make flattened content editable again.
• Workflow progress tracking — Multi-step workflows now show task planning and durable progress as they run.
• More accurate text editing — Text tools measure with the real renderer for pixel-accurate sizing, and edits resolve as one clear intent.
• Curated starter images — Image-input workflows ship with default images to get going faster.
• Get Started upgrades — MCP docs entry points, a Help link in the header, and design/motion entry points that drop you straight into focus mode.
• Beta badge with feedback — The dashboard shows a Beta badge; click it to leave feedback.
• Unified search + sticky dashboard topbar — Boards and Get Started share one search, and the topbar stays put as you scroll.
• Export confirmation step — Exporting asks you to confirm before encoding starts.

Fixed

• Video export no longer fails silently — errors now surface instead of leaving you guessing.
• Video track edits trim in place and preserve their offset instead of re-timing the whole track.
• Dragging one video clip no longer drags every track that shares its curve.
• Auto-placement accounts for masks and cropped bounds, so new content lands where expected.
• Fixed several undo edge cases (scale changes, shifted text edits, delete-only actions) that left phantom history.
• The camera can now zoom all the way out to 100%.
• Figma imports with custom fonts load correctly.
• Pasting an animated node no longer leaves ghost frames behind.
• Fixed a texture-upload edge case that could freeze the canvas session, and the agent crash-looping on large canvases.
• Reference-image generation now uses all your attached images, not a subset.
• Video generation no longer uses the wrong source image, and frame drift in generated video and animations is fixed.
• Effects and vector-path clears now sync to collaborators in real time.
• Board grid previews no longer show squashed, cropped, or stray-layer thumbnails.
• Long-form agent responses (like documents) now consistently appear as a card.
• Plus a number of smaller fixes across canvas, chat, and generation.

July 21, 2026

New

• Color grading tool — The agent can now apply color grades directly to your images and video.
• Segment & unflatten tools — New canvas tools for splitting composited layers apart and pulling objects out of flattened images; segmentation can create results in the same frame or as sibling nodes.
• Pentagon shape tool — Added to the shape toolset alongside rectangle, ellipse, and triangle.
• Right-click board menu — Right-click anywhere on the board for quick actions.
• Focus mode redesign — A decluttered focus mode with an overhauled animation editor, running on the new sync engine.
• Refined selection handles — Figma-style outlined corner handles for selection, plus dedicated corner-radius drag handles on rectangles.
• Agent confirms before gated actions — The agent now explicitly checks in with you before proceeding on workflows that require your input.
• Agent-authored animation tracks — Animations the agent creates now layer additively on top of your existing motion instead of overriding it.

Fixed

• Camera-fit on the canvas behaves correctly, and rotation now pivots around the node's center with rotation and shadows reported accurately.
• New empty frames use the correct fill color instead of appearing blank or white.
• Chat suggestion answers now wrap instead of getting truncated.
• Hex color inputs select all text on focus, and the color popover picker matches the fills picker design.
• The "Rotate by" preset nudges in 15° increments with Shift+Arrow.
• Brand fonts resolve correctly by their canonical family, and brand-uploaded fonts apply with the right scope.
• Text nodes no longer incorrectly appear in the layers panel.
• Cmd+C / Cmd+V copy-paste writes to the OS clipboard again.
• Brand Hub hides the remove control while a source is still processing, blocks deletion mid-process, and unfolds deleted sources from brand.md correctly.
• Canvas image imports (drag-and-drop, replace, paste) are restricted to formats the renderer can decode, and images respect EXIF orientation.
• Drag-and-drop no longer sometimes creates a new node instead of filling the background.
• Vector mask clipping, a mask-group calculation error, and dropped shadows on suppressed video fills are all fixed.
• The agent no longer doubles animation tracks, with clone vs. copy-animation behavior clarified.
• In-place image-to-video edits write to the correct node, and video fill "cover" fit mode matches across creation paths.
• Storyboard generation uses your selected canvas images instead of ignoring them.
• Agent-driven text edits preserve formatting fidelity (decorations, alpha, span styles), and the agent reads a selected text node's actual content before acting on it.
• The agent updates the focused frame in place instead of duplicating it.
• Clearer error messages when a generation is blocked for human-likeness concerns or times out.

July 9, 2026

New

• Tables in documents — You can now insert and edit tables directly in the document editor.
• Text autosizing — Text boxes automatically resize to fit their content, with an optional max font size cap.
• Additive box select — Hold shift while box-selecting to add to your existing selection instead of replacing it.
• Agent auto-repairs image prompts — When a generation prompt is malformed, the agent uses GPT-4o vision review to fix it automatically.
• Smarter ad generation reference — Ad generation reuses your selected imagery, video, or composite design as its reference instead of starting from scratch.
• Workflows respect your brand and direction — System workflows use your brand.md and stated direction instead of falling back to invented values.
• Combined background + animation on one frame — The agent can generate an image-to-video background and apply a copied animation to the same frame in one step.
• Smarter layer flattening for image-to-image — When you clearly mean to send your whole composition, the agent flattens all layers before running image-to-image generation.
• Agent reads animation duration automatically — The agent reads an animation's actual duration instead of asking you for it.

Fixed

• The dashboard board filter shows correct ownership and results, and falls back to "All boards" for users who don't own one.
• Minimap, HTML overlay, and camera-padding are now consistent between the web and native canvas.
• Pasted animations keep their merge semantics, and Figma paste waits for fonts to finish loading.
• Autoplacement no longer resets rotation and scale when placing frames onto the canvas.
• Loading progress indicators no longer overshoot 100% or stall near completion.
• Seedance 2.0 model naming is consistent across image-to-video, first-frame-to-video, and reference-to-video generation.
• Agent-authored animations no longer autoplay by default.
• In-place video generation no longer references the wrong, unanimated clone of an element.
• The agent no longer reports a frame as empty when it has content.
• Replacing one element's content no longer wipes out an animation kept on other elements.
• Newly inserted videos get a default timeline track.
• Text document cards derive their title from the document's content instead of showing a blank placeholder.

July 6, 2026

New

• Timeline redesign — The animation timeline got a major UI/UX overhaul, closer to modern timeline-editor conventions.
• Segmentation decal overlays — Composite decals and masks onto your canvas without altering the underlying document.
• Improved object segmentation — Object detection and removal now uses SAM 3.1 for cleaner masks.
• Coming-soon model gating — Locked models show a reusable "coming soon" state instead of disappearing from the picker.
• Default frame background — Empty frames default to a neutral gray background instead of blank or transparent.
• Brand Hub source tracking — Assets promoted to your brand are automatically registered as a brand source.
• Refreshed cookie consent banner — The consent banner and settings drawer now match the rest of the product's design language.

Fixed

• Brand Hub and sidebar issues fixed: scroll fade rendering, missing source titles, and truncated upload file names.
• The board view toggle no longer conflicts with open overlays and dialogs.
• The agent explains what went wrong when a workflow's JSON flag value is invalid, instead of a generic error.
• Positionless generated content defers correctly to automatic layout instead of picking up a stray transform.
• URL-backed SVG assets resize by true size changes instead of distorted scale transforms.
• Raised the agent's workflow description length limit and cleaned up existing descriptions for clarity.

July 2, 2026

New

• New generation models — Krea 2 and Seedance 2.0 Mini join the model picker (Reve has been removed).
• Tileable pattern tool — Generate seamless repeating patterns.
• Image export upgrades — Pick actual-size or 2x resolution from a grid picker; video exports skip the extra confirmation dialog.
• Recommended actions & workflows — A Recommended section surfaces top picks in your action and workflow lists, and workflow covers show real media instead of placeholders.
• Selection toolbar shows the generation prompt — See the prompt that created a node right from the selection toolbar.
• Assets referenced by name in chat — The agent calls out your assets by their friendly label instead of a raw ID.
• Agent prompt repair — A dedicated subagent iteratively improves your image and video generation prompts.
• Live peer cursors on canvas — See collaborators' cursors and content-loading indicators in real time.
• Agent thinking status — See a live status update while the agent works through a task.
• Open a board in a new tab — Cmd/Ctrl+click board creation opens the new board in a new tab.

Fixed

• The credit depletion banner shows correctly when the balance marker arrives mid-stream or a generation fails, and large balances display compactly.
• The agent panel and view toggle no longer disappear behind full-screen overlays on the board.
• Video nodes mute on hover and only play audio during actual playback, fixing stray audio leaks.
• Video exports and chat-input aspect-ratio pill handling fixed.
• The @ mention menu in chat opens immediately instead of after two characters.
• Library search matches workflows by tags, keywords, and category, and covers play video previews on hover instead of autoplaying.
• Brand Hub's source list refreshes live while a fold is in progress, with a proper loading state while brand.md is generating, and uploads are capped at 40 per batch.
• Font weight selection and mapping widened so more weights resolve correctly.
• Fixed a runaway resize-observer loop in chat on fractional-DPR displays.
• Canvas shapes render colored, blurred, and multiple shadows, dashed strokes, and per-fill blend modes correctly.
• Canvas group sizing and bounds issues fixed so the agent measures and scales groups accurately.
• Significant performance improvements to canvas animation by cutting unnecessary re-renders.

Experimental

• Brand Hub — A brand.md knowledge hub with a sources panel for uploading and grounding brand assets. Available for early testing on request.

WIP

• Canvas engine rebuild — A major overhaul of the underlying canvas engine.
• Animation & timeline overhaul — A new easing library and animation model building toward a future timeline redesign.

June 24, 2026

New

• Unified agent action entry points — Quick actions are consolidated with a Recommended section and selection-toolbar prompts, so the right agent action is easier to find.
• Workflow search hub redesign — Get Started now has a searchable workflow hub and a new inspirations page in place of the old all-workflows list.
• Agent reads your own social posts — The agent can pull your first-party posts, reels, and shorts as context for a request.
• Agent saves web & social finds as assets — Anything the agent finds while browsing can be pulled in as a project asset for reuse.
• Storyboard skill — Drop in a storyboard and the agent automatically routes it through reference-based video generation.
• Smarter agent video review — Video inspection uses GPT-4o frame sampling plus Gemini 2.5 Pro, with sharper feedback on what to fix.
• Viewport-aware autoplacement — Generated content places near whatever you're currently looking at, using your selection first and falling back sensibly.
• Smarter system workflows — One-shot and multistep workflow templates auto-detect inputs, offer a single-pick UI, and can adopt existing board content.
• Brand asset coverage — Brand-aware generation draws on a wider set of your brand assets.
• Per-fill image filters — Image filters can be applied to individual fill layers instead of the whole shape.

Fixed

• The view toolbar no longer slides in when a board first loads.
• Generations that previously stalled indefinitely are caught and unstuck automatically.

WIP

• Editor sync engine rewrite — A large internal overhaul of the canvas edit/sync engine, for improved sync reliability and new editor capabilities.
• Rebuilt animation engine foundation — Early scaffolding for a new animation system architecture, ahead of user-facing animation features.

June 17, 2026

New

• Ideogram V4 — New image model for text-to-image and image-to-image generation.
• LTX 2.3 Fast video + video-to-video — Generate video faster, transform existing video with the new V2V endpoint, and upscale video with the new LTX-based upscaler.
• Agent visual inspection — The agent can visually inspect canvas content mid-task to check its own work.
• Agent document reading — The agent can read uploaded documents (PDFs, Word docs, etc.) as reference material.
• Category-based workflow library — Workflows are organized by category with a redesigned Get Started page for browsing.
• Board grid overview — Preview all the content on a board at a glance without opening it.
• Board minimap — Navigate large boards more easily with an on-canvas minimap.
• Loading placeholders — Generating content shows determinate and indeterminate progress states instead of a generic spinner.
• Text improvements — Text boxes get a more accurate caret and refined text handling on canvas.
• Focus mode drag-and-drop — Dragging content in focus mode is smoother and more reliable.

Fixed

• Video export no longer jitters or skips frames.
• Animation tracks can no longer get a negative start time.
• Group resize and scale keep children properly anchored, with the top-left corner fixed.
• The agent no longer hallucinates camera moves it didn't actually make.
• The agent writes generated copy into the text node you selected, instead of the wrong one.
• Fixed a background-shape cropping regression from object removal and detection.
• Transition ordering when joining clips in animation is fixed.
• Aspect-ratio adaptation no longer distorts videos.
• Camera and masking regressions on canvas fixed, with mask-group textures clamped to a 4K ceiling to prevent oversized-texture failures.
• The default text-to-image model is corrected after the Reve model was removed.

WIP

• Brand Hub — A brand-centered knowledge base (brand.md) that grounds generation in your brand, replacing legacy brand guidelines.
• Segmentation, decal & paint overlays — Composable segmentation, decal, and paint-overlay tooling for canvas.

June 4, 2026

New

• Image Styles library — A curated styles library you can apply to your images.
• Board zoom toolbar — A small +/− control sits bottom-right of every board; hover the percentage to reset to 100%.
• New Kyoso-curated workflows — Editorial try-on, product 360°, product shots, storytelling video, storyboards, and more.
• Brand-scoped workflows — Save and run workflows scoped to a specific brand, so each brand keeps its own set of reusable flows.
• Redesigned onboarding — A trimmed sign-up flow drops you straight onto a pre-filled, explorable board with a guided tour. Every frame has a one-click Recreate action, and the empty-chat welcome state is now a curated, workflow-first surface.

Fixed

• Avatar images now crop cleanly instead of stretching or squeezing.
• The agent no longer silently ignores selected nodes it can't find — it surfaces the problem instead.
• Library navigation stays usable while you're searching.

WIP

• Brand Hub — Extended brand-knowledge upload and training.
• Style Creation — A replacement for the sunsetted LoRA training.

June 1, 2026

New

• Workflows — Save your creation workflows and scenarios as reusable templates.
• Join video tracks 2.0 — Combine multiple clips into a single continuous video.
• Timeline tracks past scene duration — Tracks can extend past the scene duration, with magnet-snapping on track edges as you drag.
• Video cropping — Crop video nodes directly on the canvas.
• New AI video upscaler — Higher-quality video upscaling.
• Extract frame — Pull a still frame out of a video as a new image node.
• Extract prompt quick action — Describe the style of any image as a reusable prompt in one click.
• Text Document nodes — Create and edit rich text documents on the canvas. The agent reads and updates them, and new documents auto-place themselves.
• Libraries sidebar — Regrouped by what you're trying to do.
• Brand font families — Font variants are grouped under a single family entry.
• Per-user credit limits — Admins can set a monthly credit cap per user, bounded by the org's plan.

Fixed

• Video muting and track playback with perfectly aligned times now behave correctly.
• Replacing an animation no longer deletes existing video tracks.
• Mask rendering and logic corrected across first-load and dirty-state cases.
• Copy/paste events propagate reliably, and selection no longer triggers while box-selecting or dragging.
• Stuck loading placeholders and incorrect default zoom resolved.
• The timeline panel collapses to an empty state when there are no tracks.
• Reconnection no longer races across multiple tabs.

May 15, 2026

New

• Unified Libraries dialog — The floating left-side library toolbar and sliding panel have been replaced by a single Libraries dialog launched from the chat input. The sidebar groups your Kyoso libraries (Creative, brand, models, motion, and more) for fast access.
• Kling V3 Pro reference-to-video — Add image elements as references and Kling V3 Pro automatically switches into R2V mode. Durations 3–15s, aspect ratios 16:9 / 9:16 / 1:1, with an optional audio toggle.
• Happy Horse video model — New text-to-video and image-to-video model available from the chat input.
• Higher-quality multiscene videos — The multiscene/storyboard tool now produces noticeably better output.
• NanoBanana 2 as default for edit and combine — The default model for editing and combining images has been swapped to NanoBanana 2.
• Remove Background quick action — One-click background removal added to the agent collection and quick actions.
• Agent can inspect PDFs — The agent can analyze PDF assets — ask questions about a PDF in chat or have the agent generate content based on it.
• Agent can inspect videos — A new inspect_video tool returns a camera-motion prompt, a video-level caption, and dense per-clip event captions so the agent can reason about footage you've added to the canvas.
• Edit Video Tracks agent skill — The agent can edit video tracks directly — animation duration and existing tracks are read and updated coherently with video tracks.
• Multitrack timeline redesign — The animation timeline, track properties panel, and media controls have been brought in line with the rest of the design system.
• (Experimental) Workflows — Ask the agent to save your workflow and use it to replicate your entire flow.

Fixed

• Image Styles training has been sunsetted due to unsatisfactory quality and will be replaced with an improved solution.
• Brand fonts can now be picked from the typography panel.
• Board search now shows a proper "no boards found" empty state instead of a blank list.
• The agent panel correctly renders frame references as pills.
• Multitrack timeline track dragging now behaves correctly.
• Canvas and camera stay consistent during board resizes.
• Shape and stroke masks render correctly on first load.
• Figma imports for text and vector layers handle more edge cases correctly.
• Multiple pixel-alignment issues across the canvas have been resolved.
• Large boards open and stream faster.
• More content-moderation edge cases are now caught and surfaced cleanly.
• The multiscene tool now respects the requested aspect ratio.
• Multiplayer cursor labels now render as a colored pill badge without overlap.
• When a long prompt overflows, the chat input collapses to ~3 lines and re-expands on focus.
• Renaming a board now updates its tab title live.

April 29, 2026

New

• New chat agent — Better reasoning, more reliable tool use, and progressive reasoning tiers. Long-running operations and multi-step plans land more cleanly.
• GPT-Image-2 — Now supports image-edit and reference-image modes, joining the lineup of image models you can call from chat.
• Camera Motion and Easing libraries — Two new libraries with curated camera-motion presets and easing curves. Multi-select is supported, so you can apply a motion + easing pairing in one go.
• Expanded Transitions library — The video transitions library has grown to 24 options.
• Boards search — A search input has been added to the boards dashboard header so you can quickly find a board by name.
• Team and user management — Admins can now invite, suspend, and delete users, and manage org-level roles.
• Texture Enhancement tool — New agent tool to enhance texture detail on images.

Improved

• Agent messages — Tables render properly in agent responses. Pills now render for tokens, tools, and frames — click them to re-insert or jump to the node. Right-to-left scripts and mixed-script text render correctly. Status messages are interleaved chronologically with agent messages so the timeline reads naturally.

Fixed

• The boards filter (e.g. "My boards") is preserved when creating a new board or switching filters.
• Chat input preserves in-progress text when pressing the up arrow, and shows a clearer file indicator on attachment thumbnails.
• Agent panel scroll no longer clamps as streamed content grows.
• Light-mode polish on the context toolbar and hover states.
• Pasting a video node over an existing node no longer breaks layout.
• Markdown element styles and text formatting in chat improved.
• Video export fix.

April 20, 2026

New

• Seedance 2.0 — New video model supporting text-to-video and image-to-video generation.
• Kling 3.0 Pro — New video model with higher-quality output.
• Video Upscale quick action — One-click video upscaling from the quick-actions toolbar.
• Video mute — Mute and unmute individual video nodes directly on the canvas.
• PDF upload in chat — Upload PDFs into the chat panel for your agent to reference.
• Video mockup templates — 13 new templates to help you start video generations faster.
• Help Center — A new help center with 70+ articles covering getting started, concepts, guides, reference docs, account management, and FAQ.

Improved

• Video timeline — Redesigned for a smoother video-track editing experience.
• Sign-in and sign-up pages — Visually refreshed.
• Top bar — The profile dropdown is now merged into the collaboration presence avatar.
• Refined copy and updated content across the platform.

Fixed

• Chat no longer jumps to the wrong scroll position when loading history or receiving agent responses.
• Animation easing curves now display correctly in the timeline panel.
• Playback timestamp toolbar and animation export work as expected.
• Focus mode correctly handles resize and exit.
• Shadows render correctly on nested elements.
• Multiplayer cursors and selection visuals no longer go stale.
• Text formatting no longer incorrectly applies to non-text elements.
• Pasted grouped content now syncs correctly across collaborators.

April 2, 2026

New

• All tools on the canvas — Every tool is now surfaced directly on the canvas and reachable from the agent. Describe a full workflow in natural language and the agent will execute it end-to-end.
• Animation variations — The animation tool now generates multiple variations from a single prompt so you can explore options or converge on the right motion.
• Copy animations between nodes — Select two nodes and ask the agent to copy the animation from one to the other.

Improved

• Animation timeline — More stable and follows instructions more reliably.
• Selective animation edits — Prompts like "keep the animation on the text and replace the animation on the image" or "keep only the animation on the text" now work as expected.

Fixed

• Deleting a selected animation no longer removes the entire animation.