Montage Editor, Prompt Enhance, and Seedance 2 Support: Changelog March 28 – April 3, 2026

A big week. The Montage video editor moved from branch to production. A new LLM-powered Prompt Enhance button replaced the old toggle. Seedance 2 multi-endpoint video support landed. Team invites no longer redirect new users to a separate signup page. Here is everything that shipped March 28 through April 3.

Montage Video Editor Is Now in Production

The Montage editor merged to main on March 31 via PR #293. It is a storyboard-based video composition tool built into Platform Studio. You generate or import individual video clips, arrange them on a timeline, and export a single stitched video.

The first frame of each clip shows the actual trimIn offset, not frame 0, so thumbnails match what will appear in the final video. Export is disabled until all clips are in READY state.

Prompt Enhance Button: Now Powered by an LLM

The old feature-flag-controlled enhance toggle is gone. In its place: a sparkle button directly in the prompt input. Click it and the current prompt gets sent to an LLM that rewrites it with model-specific context.

The enhance is context-aware. For Seedance 2, the system prompt enforces multi-shot structure: three shots by default, evenly divided durations, one motion verb per shot, camera specified as [move+speed+stability], secondary physics required (fabric, hair, drape), and a Constraints line for cross-shot consistency. Because Seedance 2 does not accept face reference photos, the LLM now describes the model's appearance (age, build, hair, skin tone) so the output is consistent across shots.

The enhance button shows a shimmer loading state while the LLM responds. It is wired into all PromptInput locations: Studio, Batch Generation, Image Edit, Video Generation, and Storyboard.

Inline Prompt Expansion

The expand button on PromptInput no longer opens a separate modal. It now expands the textarea inline as an absolute overlay: minimum 200px tall, maximum 400px, click outside to collapse. The deleted ExpandedPromptModal and its CSS are gone.

Video models without first-frame support now require a prompt before the Generate button activates. Those models also default to always-expanded mode.

Video Generation: Multi-Endpoint Model Support

Seedance 2 uses multiple inference endpoints, and each endpoint accepts different inputs. The video generation UI now adapts to this.

A frontend model access gate also shipped alongside this. It restricts model visibility by company type, email domain, and billing country. It is currently a soft-launch gate with placeholder config for Seedance 2. Backend enforcement is coming separately.

Team Invites: One Step Instead of Six

Previously, a new user receiving a team invite with no existing account had to leave the invite page, sign up, verify email, then return. Six steps.

The invite page now handles account creation directly. New users fill in first name, last name, and password without leaving the page. Google OAuth signup is also available inline. The page shows the reserved email masked, fetches the password policy from the API, and creates the account and accepts the invite in a single submission. This shipped March 31 via UWE-276.

Batch Generation: Reference Images Now Supported

Batch prompt inputs now support attaching reference images. The option appears only when the selected AI model supports image references. The reference URLs pass through to the batch request payload for both clothing and outfit item types.

The Asset Picker itself also got a design system migration this week. The old MUI Modal, Grow, and useMediaQuery components are replaced with DSModal, DSTabs (with sliding indicator and responsive overflow handling), DSInput, DSSkeleton, DSCheckbox, DSFileUpload, and DSChip. The clothing sidebar in the picker is now collapsible: hover to expand or collapse, with animation. Search clears on collapse if a query was active.

Fix: Results from Agent and API Now Show Up

The results page was silently filtering to only show generations with origin=web. Anyone who generated via the agent or API saw an empty page. That filter is removed. All results show regardless of origin.

Report Issue Button with Auto-Screenshot

The side nav now has a Report Issue button. Clicking it captures an html2canvas screenshot of the current viewport and attaches it automatically to the support form. Object URLs are revoked on close to prevent memory leaks, and a double-screenshot guard prevents duplicate uploads.

Platform Stability Fixes

Questions or feedback? Reach out. We read everything.