Changelog

Generated images and videos in Agents now have quick-action controls right on them — download, copy link, open in a new tab, and click an image to view it full-size. Auto-captions from text-to-speech are now timed word-by-word, so captions appear in sync with the spoken audio. Animated graphics on the timeline also no longer display at incorrect dimensions.

Chat history search in the Agents sidebar is faster and easier to use — particularly noticeable when scrolling through a long backlog of past conversations.

A generation-heavy release for Agents. You can now ask the assistant to generate images from a text description, optionally using reference images or frames from your own video as guidance; short video clips work the same way, with optional reference images for style and content. A new frame extraction tool pulls individual frames from your videos — useful for freeze frames or reference images. The assistant shows estimated costs and asks for approval before generating images or videos, and cost estimates themselves are more accurate. SVG graphics generation got a small quality bump at a slightly higher cost.

Workflows added shared templates (others can use a template without it being publicly listed) and starter variables (pass custom values at the start of a run). Shared templates can’t chain to private templates, which prevents broken workflows. Editor playback synchronization also handles timing drift and image clips more reliably.

In Agents, you can now add video clips from direct URLs to your timeline — pull video or image content from external sources without first importing it into your library. URL-based clips behave like library clips: trim them, reposition them, adjust their audio, add keyframe animations. Image URLs can similarly be placed on the timeline as still clips. One caveat: the assistant’s understanding of direct URL clips is limited compared to library content. A bug where some image clips were incorrectly treated as video is fixed, and the settings modal adapts better to smaller screens.

Animated graphics can now live directly on the Agents timeline — move, resize, and trim them just like regular video clips. They play back in the editor preview so you can see them in context before rendering, and they’re included in both rendered output and Premiere Pro exports. The assistant can also add animated graphics from a URL directly to your timeline.

A full feedback layer for rendered videos in Agents. Draw on the video, add comments at specific timestamps, and send the whole package directly to the assistant. Time inputs on feedback markers are editable by typing, scrolling, or arrow keys for precise adjustments. Hovering over a marker shows a preview of the comment and drawing, and clicking an existing marker lets you edit it, change its time range, or delete it. Feedback comments support rich text formatting.

Clicking the video pauses or resumes playback, the timeline scrub bar is taller for easier seeking, and drawings now come with descriptive context that helps the assistant better understand your annotations.

The app’s table view now includes a YouTube ID column for imported videos, and you can filter by full URLs, IDs, or partial matches.

In Agents, undo and redo now support up to 50 steps and are more efficient, and the video player initializes more reliably.

When rendering wide videos to portrait (or portrait to landscape), Agents can now automatically split-screen the output — top/bottom for portrait targets, left/right for landscape from portrait. The feature is applied at render time; preview in the editor is coming later.

In the app, the “Autoplay moments” toggle now works correctly — turning it off actually stops moments from playing automatically.

The external plugins section in Agents got a proper home: a suggested servers list (including Epidemic Sound) you can add in one click, toggles for enabling or disabling individual custom servers without removing them, per-server tool listings with descriptions on hover, sign-in support for servers that require OAuth, and a refresh button for a server’s tools. In the editor, overlays preload ahead of time for smoother transitions during playback, images added as overlays display correctly in the video player, and video overlays play correctly for regular (non-streaming) video files.

In the app, audio from search results only plays on hover now instead of unexpectedly. Libraries with a mix of YouTube imports and uploads show both the “Published” and “Added” date columns in table view. The grid layout uses a uniform card size based on the majority orientation in your library, with better scaling on large screens, and video thumbnails now show the full frame instead of cropping.

In the app, clicking the logo now opens the changelog — a visible reminder of what’s shipping. Re-uploading a file that previously failed also no longer gets incorrectly flagged as a duplicate.

In Agents, the welcome screen added personalized greetings based on the time of day, and your saved recipes appear there for quick access. Questions from the assistant now support rich text formatting, and when multiple questions are asked, you can skip ahead with a “Next” button instead of having to answer everything before submitting. Media thumbnails show a placeholder when an image fails to load instead of breaking. In the editor, audio playback is smoother when multiple clips are on the timeline, and large project downloads no longer time out.