Changelog

📦 New

• Multiplayer editing — invite someone to a workflow and build together, live. You see their cursor with their name, their avatar in the top bar, and a colored outline plus name tag on whatever node they have selected, so you always know who is working where. Edits sync instantly, and dragging stays smooth even with several people in the room.
• Comments on the canvas — press C (or hit the comment tool) and click anywhere to leave a note. Comments pin to a node and ride along when it moves, threads support replies and resolve, hovering a pin shows a quick preview, and a comments panel lists every open thread with click-to-jump. New comments appear for everyone in the room the moment they are posted. @mentions are highlighted, and a thread auto-resolves if its node gets deleted.
• Team chat with your AI agent — the AI assistant now has a Team tab: one shared conversation per workflow that every invited collaborator can read and continue. When the agent is building, everyone sees it appear on the canvas as its own presence, with an outline on the node it is creating. Your personal chat stays private.
• Conflict resolution, the sane way — if a workflow is changed outside your tab (a teammate, or your own agent via the API) while you have unsaved edits, you now get a clear choice: load their version or keep yours. No more mystery "unsaved changes" states.
• Share with real permissions — share a workflow as view-only or editable, invite specific people, and revoke access at any time. Revocation is immediate: a removed collaborator is disconnected from the live session on the spot, not at their next refresh.

🧠 Improved

• The "save error" is dead — we traced every path that could knock a workflow's save state out of sync (there were several, including an invisible one triggered by link previews) and closed the whole class permanently, with an automated gate so it cannot come back. Autosave now just works, solo or in a room.
• Editor stability — fixed a canvas crash, a stale-chunk load error after deploys, and a case where selection outlines lagged behind a node while dragging.
• Long renders are more durable — video generation now runs on durable infrastructure that survives timeouts and closed browser tabs, and recovers jobs that would previously get stuck.

🔄 Changed

• Security hardening for shared workflows — collaboration sessions use signed, short-lived access tokens, view-only collaborators are enforced server-side, and commenting on a shared workflow requires a real invite (a public link alone stays read-only).

📦 New

• Sound effects, two ways — ElevenLabs Sound Effects (the best text-to-SFX there is — type "soft notification pop" and get it) and MMAudio, which generates realistic foley synced to a video clip (footsteps, ambience, impacts that track the motion). Drop a notification ping when a banner appears, or make silent AI b-roll sound real.
• Captions you can actually edit — correct any word and it sticks (perfect for brand names that auto-transcription mishears), and restyle every detail from the schema: active/inactive color, outline width & color, font, weight, size, letter-spacing, case, and glow. The outline keeps the highlighted word readable on any background — light screen-recording or dark hook.
• Bulk image generation (data-driven) — design one Compositor layout, add a table of rows, and get a whole set of finished, on-brand images in a single run — one template, many variations. A visual Batch panel shows live per-row previews, and a Build Data node wires a generation step straight into the batch.
• Build & render videos through the API — upload your own assets, author a video’s scenes, captions, voiceover, sound, image overlays and gradient backgrounds in JSON, dry-run a cheap preview before you pay, then render — all headlessly. Designed so workflows and AI agents can produce complete, on-brand ads programmatically.
• Affiliate program — share your referral link and earn on the signups you bring in; attribution is automatic.
• App Store end-card block — a polished download/end-card for app-promo reels with app icon, rating and call-to-action.

🧠 Improved

• Re-render just one node — change one thing without redoing everything. Re-run a single node and reuse your existing AI generations, so you can tweak the timeline, captions or styling without re-rolling (and re-paying for) the videos and images you already liked.
• When something’s wrong, you’re told exactly what — saves and renders now return precise, actionable feedback (a dropped scene, an empty caption track, a missing asset) instead of failing silently, so fixes are one step instead of ten.
• Bulletproof billing — a new credit ledger makes every charge, settlement and refund idempotent and auditable, failed runs are automatically refunded, and subscription webhooks were hardened so paying customers reliably receive their credits.
• Authoring just works — build a node from its type and the platform fills in the rest (the right inputs, the right settings panel), so workflows come out correct instead of subtly broken.

🔄 Changed

• Accurate, verified pricing — model prices are checked against each provider’s live page, so the credit estimate matches what a run actually costs.
• Pipe any fixed asset through the graph — a screen recording, a logo, or a pre-made clip now flows through a workflow like generated media via new video and audio import nodes, instead of being a hardcoded link.
• Captions stay in sync — voiceover and karaoke captions track the real audio length, and legibility is improved across light and dark scenes.

📦 New

• Draw on the canvas — a freehand brush in the Compositor lets you sketch or annotate with smooth strokes, a Figma-style toolbar, and a clear Pen / Select toggle. Wire the drawing into any image model: the new Doodle to Image template turns a rough sketch into a cinematic render in a single run.
• Krea 2 & Ideogram V4 — three Krea 2 tiers (Turbo / Medium / Large) for high-fidelity text-to-image with style references, plus Ideogram V4 for best-in-class text rendering. Both added with correct pricing and the full option set pulled straight from each model’s live schema.
• A chime when a run finishes — long renders now play a sound the moment they complete, with a settings toggle, so you can switch tabs without babysitting the canvas.

🧠 Improved

• Models stay in sync with their providers — a new daily check validates every model’s defaults and options against the provider’s live schema, so settings can’t silently fall out of date, deprecated or renamed endpoints surface loudly, and adding a new model is now a near-instant scaffold instead of hand-written config.
• Reference-to-video with a start frame — point a model at a start image plus reference images (e.g. a doodle of the motion you want) and it follows them, auto-detecting whether each connected source is an image, video, or audio.
• Top up credits without leaving the editor — the "get more credits" path now opens a real in-app top-up flow instead of dead-ending.

🔄 Changed

• Compositor polish — drawings-only compositions now show a preview thumbnail, the canvas edge stays visible against dark art, and freehand layers render correctly into both the on-canvas preview and the final image.
• Sharper pricing & previews — Seedance bills at the right per-variant rate, image-size presets match what each model actually accepts, the credit pre-flight checks the account a run truly charges, and pasting a /flow link now unfurls a per-workflow social card.