Changelog

Added

• Added a binary-upload path for ChatGPT- or agent-generated local image files, so they can become Flare assets and be inserted into the active canvas.
• Added native AI Remove and Stylize actions for selected image layers, including public style previews.
• Added MCP App support for completing an image annotation and sending the resulting request back to the Codex conversation.

Improved

• Improved annotations so arrows, text, and other marks scale naturally with both the target image and the canvas zoom, without arbitrary size ceilings.
• Improved text-effect references by rendering the actual text layer at high resolution before generation. Font choice, size, spacing, transforms, and position now remain visible to the model.
• Improved text-effect composition so results use the background image dimensions, return to the original text position, and can replace the temporary source layers in the canvas.
• Replaced generic loading states for canvas generation actions with the woven shader loading treatment used elsewhere in the editor.
• Kept temporary text-effect references outside the normal asset library while preserving their short-lived access for generation.

Fixed

• Fixed MCP media handling so agent-created images stay canonical after insertion into the iframe editor.
• Fixed image annotation saves from the MCP App and eliminated stale duplicate Codex presence entries.
• Fixed project title feedback in the MCP editor so a successful Enter-to-rename does not visually revert to the old name.
• Fixed multi-selection collaboration geometry and automatic layer naming in common canvas workflows.

Added

• Added the Flare editor as a remote Codex MCP App, so an agent can open and work in a project directly from a Codex or ChatGPT conversation.
• Added signed, app-origin editor sessions that connect to the same live project room as the browser without exposing the MCP OAuth token to the editor surface.
• Added shared project presence, allowing active editor sessions to be represented in the collaboration workspace.
• Added project-aware MCP editor controls for opening the browser project and refreshing an expired editor session.

Improved

• Improved the embedded editor lifecycle around loading, fullscreen, host resume, and session refresh, so the MCP App feels like the same project rather than a detached preview.
• Improved collaboration transport around reconnects, queued local edits, initial viewport fitting, and remote-update recovery.
• Improved the MCP editor chrome with dedicated sync feedback and a more focused tool surface.
• Improved production readiness across signed widget resources, app-origin hosting, OAuth scopes, and collaboration diagnostics.

Fixed

• Fixed cases where a Codex editor could lose pending edits or show stale project state after reconnecting.
• Fixed duplicate root layers, stalled multi-selection updates, and abandoned live-transform state in shared projects.
• Fixed editor resume and paint cases where an embedded canvas could return without visibly restoring its content.

Added

• Added a more capable image action workflow for selected layers, including annotation, crop, expansion, merge, and AI-assisted image-editing entry points.
• Added Nano Banana 2 Lite to the available generation models for fast image creation and revision workflows.
• Added text-effect generation for editable text, with preset directions and an optional custom lettering prompt.
• Added text-and-image context for effects, so a lettering request can use the surrounding image when it needs to blend into the scene.

Improved

• Improved text-effect requests to retain the selected text’s placement and visual reference, making styled lettering feel connected to the original composition instead of arriving as an unrelated image.
• Improved the result placement path so generated lettering can replace the working visual at the intended canvas position.
• Improved image annotations so their callouts stay readable as the canvas zoom changes.
• Improved image-action controls and generated-layer feedback so common edit actions are easier to discover from the canvas.

Fixed

• Fixed generated image recovery and layer-selection edge cases that could leave the wrong layer active after an AI action.
• Fixed image annotation persistence and overlay positioning during image edits.
• Fixed text-effect blend placement and kept intermediate reference media out of the normal asset library.

Added

• Added site converter tools for local image, video, and animation format conversion, with English and Chinese routes, footer entry points, and sitemap coverage.
• Added browser-side conversion support for WebP, PNG, JPEG, WebM, Animated WebP, GIF, and APNG workflows.
• Added admin audience and operations analytics so internal views can track users, workspaces, operation runs, reservations, and settled usage over time.
• Added a generated site sitemap and robots endpoint for stronger public-site indexing.
• Added expanded product overview and agent workflow docs in both English and Chinese.

Improved

• Improved the editor export pipeline by extracting shared media export helpers used by frame and sequence exports.
• Improved generation controls, image toolbar behavior, and mask controls so generated media workflows feel calmer and more predictable.
• Improved admin operations so they work globally instead of being tied to a single workspace, with more compact admin console layouts and simpler settings copy.
• Improved MCP collaboration operations by binding live browser sessions more directly to project collaboration state.
• Improved asset storage and search limits, including clearer docs and stronger worker-side enforcement.
• Improved recipe run and operation run plumbing so generated assets, billing events, and run history stay easier to inspect.

Fixed

• Fixed initial editor viewport fitting so canvas content is visible on first load.
• Fixed Shift aspect-ratio locking and clip-mask radius dragging edge cases.
• Fixed layer copy and paste semantics for nested and selected layers.
• Fixed layer downloads so exported files match the visible display size.
• Fixed media upload CORS headers for browser upload flows.
• Fixed image toolbar and shape mask alignment details, plus several editor control polish issues.

Added

• Added the public beta blog launch, blog navigation, polished post headings, and updated home/social preview assets.
• Added refreshed Flare brand assets, app icons, favicon updates, and a dedicated brand page for the logo system.
• Added production MCP rollout docs, updated Flare skill guidance, and direct MCP project creation with route-aware project URLs.
• Improved MCP-created media insertion so generated layers keep their dimensions, visibility, and annotations.
• Added the Shader tool and shader layer workflow foundation, backed by the shared shader core and shader page.
• Added beta Recipe metrics summaries and Google Analytics tracking.

Improved

• Improved image annotation and revision workflows, including clearer generated-image upload source paths and revision-first generation copy.
• Improved generated asset search metadata so Codex-created media and related fields are easier to find semantically.
• Improved shader entry points and naming so shader layers are easier to recognize in the editor and supporting pages.
• Improved editor image loading and deletion paths, including cleaner asset source presentation and bulk cleanup behavior.
• Improved collapsed editor chrome sizing, generator popover layout, and localized generator copy.
• Improved public site navigation, blog layout, Chinese launch copy, and homepage slogan language.
• Improved app environment origin handling and public CDN sync tooling.

Fixed

• Fixed delete-key handling across selected canvas layers, nested layers, active selections, and text-editing fields.
• Fixed layer and project rename fields swallowing Delete/Backspace while editing names.
• Fixed MCP-created images and imported frame text disappearing or losing expected sequence behavior.
• Fixed completed Stripe checkout reconciliation so already-paid sessions no longer stay stuck in sync.
• Fixed duplicate collaboration presence peers accumulating in the editor toolbar.
• Fixed app dark-mode projects starting with a light canvas background.
• Fixed asset delete confirmation placement and several mixed-language UI labels.

Added

• Added expanded bilingual docs covering first-canvas setup, generation and arrangement, asset folders, timeline motion, preview/export, and troubleshooting.
• Added Markdown documentation output and llms.txt endpoints so agents and search systems can read public docs more directly.
• Added bulk selection and deletion in the asset library to make generated candidates and temporary media easier to clean up.

Improved

• Improved docs site information architecture, mobile navigation, detail-page layout, and code block rendering for longer-form reading.
• Improved public site navigation and footer by hiding entries that are not yet backed by enough content.
• Improved pricing copy hierarchy across plans, top-ups, FAQ, and calls to action.
• Improved app settings navigation with a more unified sidebar background and temporarily hidden Layer Actions entry.
• Improved editor title chrome and rail menu hover feedback to reduce visual noise around the workspace.
• Improved localized UI labels across billing, the layer panel, path tools, color inputs, and lock/play tooltips.

Fixed

• Fixed crowded mobile layouts on docs index and detail pages.
• Fixed mixed-language labels, active states, and divider styling in app settings and workspace navigation.
• Fixed empty public site sections still appearing in navigation and the footer.

Added

• Added QuiverAI Arrow 1.1 and QuiverAI Arrow 1.1 Max as SVG generation models in the shared generation queue.
• Added Quiver SVG request handling, output parsing, result ingestion, and billing coverage so SVG results can reuse the existing generation and asset pipeline.
• Added MCP tool-call allowance rules by plan, while keeping binary upload, storage, AI generation, cloud rendering, and batch/API work under their own limits.
• Added plan gating for semantic asset search to prevent workspaces without the entitlement from using higher-cost search capabilities.
• Added annual billing rules and tests for monthly entitlement delivery.

Improved

• Redesigned the public pricing page with monthly and annual billing toggles, clearer plan cards, top-up content, and a more complete FAQ.
• Improved the in-app billing interval selector to make plan switching less ambiguous.
• Improved checkout layout so sign-in, payment state, and selected plan details balance better across narrow and desktop screens.
• Expanded billing and MCP docs to clarify boundaries between tool calls, generation, rendering, storage, assets, and search.

Fixed

• Fixed billing fixtures and tests so the new allowance rules are covered.
• Fixed generation result ingestion for SVG output type and media metadata handling.
• Fixed incomplete or inconsistent localized labels in plan and billing surfaces.

Added

• Added AI annotation mode for image layers, letting users mark up a target image with arrows, rectangles, circles, and text before asking an external agent to revise it.
• Added an immersive annotation view that focuses the target image, quiets the rest of the canvas, and keeps the interaction closer to native canvas drawing.
• Added a compact annotation toolbar with arrow, rectangle, ellipse, text, and common colors, defaulting to arrows for fast local change requests.
• Added undo, selection delete, and complete actions for annotations, while keeping annotation data separate from ordinary design layers.
• Added the get_image_annotation_context MCP tool, returning the target image node, asset provenance, structured annotations, suggested revised-image placement, and an optional composite preview.
• Added a normalized 0..1 coordinate contract for annotation targets, so arrow targets, region bounds, and text positions are relative to the target image instead of the current viewport.

Improved

• Annotation text can sit outside the image while arrows point into the target area, reducing visual obstruction over the image itself.
• Annotation data is organized as a task-level artifact. After a revision is generated, users can annotate the next result as a fresh task.
• Composite previews are generated from the target image and annotation data, rather than from the full viewport, reducing accidental inclusion of nearby layers or side panels.
• Agent revision flows default to the agent’s own image capability, then binary-upload the revised file and place it next to the original with insert_asset_image.
• Added a transition when entering annotation mode so the focused view feels less abrupt.

Fixed

• Fixed annotation arrow geometry where the line could extend past the arrow head.
• Fixed text input styling so it is lighter, smaller, and closer to the rest of the canvas UI.
• Fixed zoomed annotation mode cases where sidebars or normal canvas content could show through and overlap the target image.
• Fixed deletion behavior that could remove the target image instead of the selected annotation.
• Fixed annotation positioning during canvas pan, zoom, and target-image transforms.

Added

• Added create_image_upload_session so external agents can create a short-lived binary upload session before uploading locally generated image files into Assets.
• Added insert_asset_image, insert_agent_generated_image, and Codex-compatible aliases for placing images an agent already generated or obtained, without accidentally starting a Flare backend generation job.
• Added provenance for agent-generated media, including source client, generation model, generation tool, prompt, and notes, so assets keep their origin once they enter the library.
• Added insert_html, converting one HTML snippet or full HTML document into one root frame with supported text, simple boxes, and public HTTPS images as editable child layers.
• Added MCP capability discovery for generation models, motion presets, shader presets, canvas patch operations, media capabilities, and render presets, reducing reliance on stale static instructions.
• Added arrow shape insertion through MCP for agent-authored annotations, diagrams, and visual callouts.
• Added check_client_setup so agents can check the local Flare skill version before the first visible canvas write and remind users to update stale instructions.
• Added selectable MCP OAuth scopes. Recommended agent read/write scopes are selected by default, while generation and render creation scopes stay optional because they can consume usage.

Improved

• Clarified that local files should use binary upload rather than base64 or data URLs inside MCP JSON, reducing context size and truncation risk.
• Improved MCP tool descriptions around center-based canvas coordinates and default root-canvas placement for generated images.
• Expanded Flare skill and MCP docs for Codex-side image generation, visible in-app browser workflows, HTML import, asset upload, and skill update reminders.
• Added MCP plan call allowances and semantic asset search gating. OAuth handshakes, initialization, tool listing, and setup health checks do not count as normal tool calls.

Fixed

• Fixed MCP image insertion sizing and intrinsic metadata handling so images are not cropped or scaled unexpectedly.
• Fixed generated image insertion recovery paths to reduce duplicate generations, duplicate layers, and inconsistent asset state.
• Fixed collaboration bootstrap, overlapping autosaves, and stale checkpoint cases that could let remote state overwrite local edits.
• Fixed an editor chunk cycle that could break editor loading.

Added

• Added path text support so text can follow editable pen paths and survive schema export, import, playback, and frame-based animation.
• Added text placement modes closer to design tools: click to start auto-width text, drag to create fixed-width text boxes, and switch between auto width, auto height, and fixed-size text layout.
• Added layer ordering commands to canvas and layer-panel context menus, including bring forward, send backward, bring to front, and send to back.
• Added fragment shader presets and generated shader thumbnail assets, expanding the shader library with ferrofluid, aurora, chrome, silk, scan, light, prism, and plasma-style visuals.
• Added updated bilingual docs pages for canvas basics, asset folders, generation, projects, shared workspaces, billing, shortcuts, timeline motion, templates, and troubleshooting.
• Added private media access indexes and public CDN asset service support for recipe and generated media workflows.

Improved

• Improved shader preset browsing with real thumbnail previews, more stable preview asset generation, and better rendering behavior for animated shader layers.
• Improved editor transform interactions with edge-based resizing, cleaner resize handles, alt-drag duplication, temporary aspect-ratio locking with Shift, and snapping during corner and edge resizing.
• Improved text and path editing controls, including path-text toggles, type layout controls, and safer behavior when text boxes should not use path text.
• Improved layer panel and timeline labels, i18n coverage, shader icons, and context menus so newer editor surfaces avoid mixed-language UI.
• Improved MP4 and frame export sizing, H.264 encoder configuration, and generation workflow state handling.
• Improved site docs navigation, document detail layout, cookie consent handling, and brand assets for the documentation site.

Fixed

• Fixed path text playback interactions so animated text-on-path layers keep their geometry and timing during playback.
• Fixed shader layer rendering issues around animation updates, offscreen work, shadows, and cached render state.
• Fixed layer right-click instability and selection jumps when opening context menus from canvas or layer rows.
• Fixed canvas transform glitches where resizing could flicker, snap too aggressively, or stop moving after hitting guides.
• Fixed i18n gaps across admin, asset, shader, motion, timeline, and properties panels.
• Fixed generated media, asset library, and recipe media flows so private and public media URLs resolve more consistently.

Added

• Added a dedicated Shader layer type for procedural gradient artwork directly on the canvas, with animated render time, blend modes, and editable shader parameters.
• Added a shader preset browser with Plane, Orb, Flow, and Texture categories, so users can start from richer Shader Gradient-style looks instead of building every abstract background manually.
• Added new shader presets and defaults, including mesh gradients, grain textures, light leaks, ribbon flow, watercolor wash, and halftone swirl treatments.
• Added a timeline-first motion clip workflow so one layer can hold multiple preset or custom animation clips instead of being limited to a single entrance and exit preset.
• Added compact preset popovers for text and layer motion, keeping the inspector short while still supporting grouped preset browsing.
• Added a selected motion summary card with clip timing, preview, delete, and advanced actions before exposing the full keyframe editor.
• Added a reusable advanced motion editor shell for preset-backed clips, so simple preset use stays clean and custom track editing remains available when needed.
• Added a standard multi-layer motion acceptance fixture covering text, image, shape, path, and shader layers inside a frame.
• Added regression tests for motion popovers, summary-card actions, multi-clip state resolution, and schema round-tripping of frame-based motion documents.

Improved

• Improved shader editing with reusable color slots, safer color uniform defaults, categorized browsing, random palette generation, and serialization support for preset params.
• Improved timeline selection for motion clips, including playhead-aware clip selection and quieter selected states on clip rows.
• Simplified the motion inspector default state so choosing a preset no longer expands every advanced parameter by default.
• Improved preset preview consistency across text, images, shapes, and pen paths, including better handling of text decorations, image blur, crop, wipe, trace, and marching-ants style motion.
• Tuned timeline panel expand/collapse timing so side panels and timeline movement feel more connected.
• Kept custom keyframes, numeric tracks, color tracks, motion paths, and loop clips available through advanced editing instead of mixing them into the default preset flow.

Fixed

• Fixed shader layer animation state so disabling animation freezes the current rendered frame instead of jumping back to the default time.
• Fixed frame composition handling during schema import, export, and reconcile so child layer timelines and motion clips respect non-default frame durations.
• Fixed motion serialization now that explicit clips are the source of truth instead of legacy animation slots.
• Fixed cases where image, shape, path, and text preset previews could appear identical or drift away from actual canvas playback.
• Fixed motion parameter controls and selected clip styling so they align with the shared editor panel system.

Added

• Added live motion preset thumbnails so animation choices can be previewed directly in the picker.
• Added grouped preset browsing for text and layer animations, making larger preset sets easier to scan.
• Added media blur strength controls for Focus, Defocus, and Depth Pulse presets.
• Added a shared color picker for text Highlight animation color and custom motion color keyframes.
• Added hover-aware generated-layer parameter affordances that scale to the selected image layer.
• Added pointer-lock support to scrubbable number fields so drag adjustments can continue when the pointer reaches the screen edge.

Improved

• Improved text preset previews so character-level timing, text styling, underline, strikethrough, and highlight color better match the canvas.
• Improved shape, path, and image previews so thumbnails use the selected layer appearance instead of generic placeholders.
• Made wipe, crop, trace, marching ants, slide, rise, and drop previews more visually distinct and closer to their real output.
• Tuned selected preset states and motion parameter controls to stay consistent with the editor panel system.
• Refined generated image layer quick actions with transform-based positioning, tighter overflow menus, and less intrusive generated-parameter triggers.
• Hid image layer quick actions and generated-parameter triggers while panning, zooming, or transforming the layer, then restored them after the canvas settles.

Fixed

• Fixed saved canvas snapshots so active frame or sequence playback previews are restored before serialization.
• Fixed a case where hidden playback-preview state could leak into exported schema visibility.

Added

• Added asset folders so workspace assets can be organized beyond source and media type.
• Added smart asset views for recent, generated, uploaded, unfiled, image, video, and audio assets.
• Added batch move actions for selected assets, including moving items back to unfiled.
• Added folder-aware asset loading and search, so folder filters work with recent assets and keyword results.

Improved

• Simplified the editor asset panel so search stays primary and advanced filters stay collapsed until needed.
• Tuned asset selection visuals to feel lighter and closer to the editor panel system.

Added

• Added structured Content Collections for Blog, Docs, and Changelog.
• Added dedicated docs routes for English and Chinese.
• Added guides for MCP setup, HTML paste, and Figma paste.

Improved

• Unified resource page content loading so future content can be maintained as Markdown.

Added

• Connected the mobile catalog to public recipes.
• Enabled mobile recipe runs with uploaded media.
• Merged profile and creation history into the mobile experience.

Improved

• Unified mobile creation entrypoints and polished native navigation.

Added

• Added public recipe slugs, site-hosted recipe runs, and the community recipe page.
• Added uploaded recipe inputs, run output handling, remix tracking, sorting, and reward claim flows.
• Added multi-step recipe workflows and board recipe runs.
• Added the first native mobile app scaffold and sign-in flow.

Improved

• Made public recipe publishing and running more reliable across site and app surfaces.

Added

• Added character-level text animations, spring easing, color and skew tracks, trim path animation, and scene transitions.
• Added the motion curve editor, keyframe dots, draggable Bezier handles, and one-click easing presets.
• Added reusable motion clip presets and the first AI recipe foundation.

Improved

• Added public recipes, recipe output tracking, and i18n coverage around recipe flows.

Added

• Added collaborative presence controls and editor sharing controls.
• Added a canvas minimap.
• Added OAuth-backed MCP server access for project and canvas integrations.

Improved

• Improved collaboration socket recovery and live transform syncing while following collaborators.

Added

• Added account management and transactional email templates.
• Added app and editor i18n support.
• Added app theme mode support and avatar image upload.

Improved

• Polished editor controls, asset terminology, language settings, and dashboard interactions.

Added

• Added persisted collaboration sync acknowledgements.
• Persisted project drafts across browser sessions.
• Added workspace invites and email delivery for invites.

Improved

• Stabilized auth bootstrap, editor document loading, and resize behavior.

Added

• Added D1-backed metadata tables and repositories.
• Added operations recovery tooling, request observability, rate limiting, and disaster recovery docs.
• Added cursor pagination and cleanup workflows for workspace listings and stored resources.

Changed

• Moved asset, project, generation, and render job metadata out of legacy storage paths.

Added

• Added collaboration room sync, delta sync, text sync, structural ordering, and advanced collaboration sync.
• Added hybrid asset search.
• Resolved render exports from canonical collaboration state.

Changed

• Canonicalized project documents through the collaboration room.

Added

• Added billing schema, service layer, and job charging for generation and render work.
• Added Stripe checkout and webhook flows.
• Added pricing page, catalog-driven checkout, quote previews, and insufficient balance states.

Improved

• Added reconciliation and admin adjustment paths for billing operations.

Added

• Added generation job contracts, persistence, and editor generator UI.
• Added generated output ingestion into account assets.
• Added generation model capability metadata and model-specific options.
• Added video generation support and video frame/resolution capabilities.

Improved

• Added generator safety and reliability controls.

Added

• Added the Astro public site scaffold and product workspace layout.
• Added auth, account, team member management, and persisted project routes.
• Added shared assets, brand kits, and queue-backed render runner infrastructure.

Changed

• Migrated the app into the product workspace and hardened render authentication.

Added

• Added frame composition metadata, playback preview, preset animations, and MP4 export.
• Added motion clip authoring and timeline subtracks.
• Added sequence creation, sequence playback, thumbnails, scrubber, and export flow.
• Added captions, project audio, and audio-aware sequence export.
• Added render job API, lifecycle routes, and a headless render job runner.

Added

• Added auto-layout groups, nested auto layout, fixed/flex sizing, and overflow controls.
• Added snapping, measurement overlays, guide toggles, and transactional undo/redo history.
• Added path drawing, Bezier authoring, stroke controls, reverse path direction, and path mode conversion.
• Added multi-selection and frame reparenting stability improvements.

Added

• Added the initial editor project, core components, and build configuration.
• Added draggable panels, image and SVG library panels, transform controls, and typography controls.
• Added canvas object placement, layer drag/drop, context menus, and the first SVG path editing workflow.