AI Shorts turns a long recording into short-form cuts: it transcribes the speech, asks a chat model to pick the strongest self-contained moments, and hands them to you as ready-to-insert 9:16 candidates. AI Shorts is a PRO feature.
Run it
Select a video layer and open Smart AI → AI Shorts… in the timeline toolbar:
- Setup — pick the detection model and options, including optional captions (off by default, animated with any preset from the captions picker and voice-synced word by word) and a Motion blur toggle that adds the stock Motion Blur layer effect to every inserted short so animated crops smear naturally.
- Progress — transcription and detection run as a background job with per-phase progress, a live log and Cancel. Speech beyond the transcription ceiling (~100 minutes) is transcribed in slices and stitched; transcripts are cached per file, so retrying with a different model re-ranks without transcribing again.
- Results — a wall of 9:16 candidates, each a real player looping only inside its own range, with the exact transcript excerpt, an AI score, a hook and a one-line rationale. Details and Crop opens a full-height player with sound, word-aligned in/out timecodes and a ribbon showing where the moment falls in the recording.
Crop
Each short has a CROP panel:
- Auto — bakes a centred fill crop so the insert matches the 9:16 preview exactly.
- Custom — the poster becomes the drag surface: drag to reframe, mouse-wheel to scale, and a keyframe lane below writes keys at the playhead. Drag keys to retime; the chips between keys pick each segment’s ease (Hold, Linear, Ease In/Out, Smooth, Overshoot, the AE curves, and more).
On insert the crop bakes as ordinary transform keyframes against the destination composition — the source is never re-rendered.
Insert
Tick the moments you want and send them to the current timeline (a new “AI Shorts” layer above the source, removable with one Undo), a new 1080×1920 project, or a single nested Sequence. Every short is just a trim window on the original clip, so eight shorts off one recording still reference one video.
The analysis is saved on the layer — reopening the wizard offers Use saved, and the AI Data panel lists it (removable).
Leave a Reply