Make a reel from a video

Give Kinetic Scenes a long video and a length; it finds the moments worth keeping and cuts them together into a reel. Every part stays an ordinary footage clip, so the result is a starting point you edit, not a finished render you have to accept or redo.

How to use it

1. Tools › Make a reel from a video…
2. Choose a video. Anything the browser can play: a recorded talk, a product demo, a screen recording, phone footage.
3. Pick the length (15 to 90 seconds), the shape (vertical 9:16 for Shorts, Reels and TikTok, square, portrait 4:5 or landscape), and whether each part gets a title.
4. Make the reel. The AI Dev Kit on your computer watches the video (a few seconds per minute of video; longer when it also transcribes the speech). The reel replaces the current scene; Undo brings your scene back.

Then finish it like any scene: add music from Sounds, use Reframe to follow the face on wide footage in a tall reel, swap a part for a better moment by changing its start in the footage (the clip's offset), or trim one.

What it looks for

Each second of the video is scored on:

	What it means	Why
Sharp	detail in the picture	blurry and out-of-focus stretches lose
Steady	little change inside a shot	shaky or panning moments lose
Well lit	neither dark nor blown out
Talked about	words being said	a reel keeps what the speaker says; needs speech recognition (below)

• Shots: the cuts in the video are found first, and a part never crosses one (no half a shot of something else).
• Picking: the best-scoring stretches are taken first, kept apart, and spread over the whole video, then put back in the order they happened.
• On the beat: with music, each part lasts a whole number of beats, so the cuts land on the beat.
• Each part is a footage clip with a hard cut and a slow push-in (a keyframed 6% zoom) so held shots don't look frozen. Its name says where it came from and why it was picked (Part 2 (10-13.21 s: sharp, steady, talked about)).
• Titles come from the first few words said in the part. They are a first draft: rewrite them. Turn them off for a video that already has words on screen.

Limits

• Speech needs speech recognition installed in the AI Dev Kit (select any audio clip, Words, Install). Without it the picture alone decides, and parts have no titles.
• It judges how a moment looks and whether something is said; it does not understand what is shown (a product being held up, a reveal). Check the parts and swap any that miss the point.
• The video is sent to the AI Dev Kit on your computer, never uploaded. Very large files (over about 500 MB) are best cut down first.

For AI agents

studio-cli highlights talk.mp4 --length 30 --format 9x16 --music track.mp3 --out reel.json

--part 3 sets the seconds per part, --no-titles leaves titles out, --title names the scene. The MCP tool is video_highlights. Hand the scene over for preview; render only when asked.