The Scene extractor turns a video on the canvas into still frames — one per shot — so you can review, annotate, or storyboard the footage without scrubbing a timeline. It detects shot changes automatically and lays the results out for you as canvas frames.
Where to find it
Right-click a video element (or open its ⋯ more-actions menu) and pick one of:
- Extract scene frames
- Extract first-last scene frames
- Prepare template for Storyboard
The video must already be downloaded locally — these actions analyse the file on disk, not a remote URL. If the clip hasn’t been downloaded yet you’ll see “Scene frames need a downloaded local video.”
Scene detection runs on the backend (FFmpeg shot-change detection) and is fully automatic — there’s nothing to configure. While it works you’ll see “Detecting scenes and extracting…”, and when it finishes a notice reports how many frames came from how many scenes, e.g. “Added 12 frame(s) from 6 scene(s).”
Extract scene frames
Detects every shot change and grabs one representative frame per scene (taken from the middle of each shot). The frames are placed inside a new “Scene frames of …” frame, arranged in a 4-column grid directly below the source video. Each thumbnail is labelled with the video title and its timecode.
Use this for a quick contact sheet of a clip — a fast way to see everything that happens in a long video at a glance.
Extract first-last scene frames
Detects the same shot changes but grabs two frames per scene — the start and the end of each shot. They’re laid out in a 2-column grid (start | end), one row per scene, again inside a “Scene frames of …” frame below the video.
Use this when the motion within a shot matters — comparing the first and last frame of each scene shows you how the camera or subject moved across the shot.
Prepare template for Storyboard
Builds a complete storyboard scaffold from the video, ready for the Board Agent to describe:
- An outer “Storyboard for …” frame wraps the whole thing.
- For each detected scene, a “Scene N” frame holds three thumbnails — first · central · last — so the shot reads as a beginning-middle-end strip.
- Below each strip, a dark Timecode note is pre-filled with the scene number and its time range (e.g. “Scene 1 · 0:00 → 0:15”), leaving room for a description.
- A green agent task note sits beneath the storyboard with instructions for the Board Agent to study each scene’s frames and append a description to that scene’s note.
When it finishes you’ll see “Storyboard template ready (N scene(s)). Run the Board Agent to fill descriptions.” The template itself runs no AI — it’s the layout. Run the Board Agent afterwards to fill in the scene descriptions automatically.
Leave a Reply