Reel editing with the Assistant
The Reel Editor is a built-in Assistant persona that cuts a short music-driven reel from a bin of clips and a song. It works the way editors have been cutting reels with an AI by script, only inside Kitlig: it never watches the footage, it measures it and looks at still frames, lays the shots on the beat grid, renders once, and checks its own work from a strip of frames. You give notes by time and clip name; each revision is a new version of the sequence.
Before you start
- Put the footage in a bin (Video ▸ Edit ▸ Media Bin ▸ New bin, then drag clips onto the chip). A name like "Food" or "DJ" is what you will say in chat.
- Import the song as an audio asset.
- Open or create the sequence the reel should live in. A fresh one is cleanest; the agent can also make one.
Starting
In Video ▸ Edit, click Reel on the toolbar. The Assistant opens on a new session with the Reel Editor picked and a starter message in the input that lists your bins. Fill in the bin and the song, say roughly how long and which shape (vertical 1080×1920 for Instagram and TikTok), and send.
You can also pick Reel Editor in the Assistant's Agent menu and just talk to it.
What it does
- Sees the footage. It starts from what the clips already say about themselves: the description, tags and timestamped moments written by Analyze with AI, and describes any clip that has none (one paid call each). Then it reads every clip in the bin: size, frame rate, length, and a camera-shake score. It also makes contact sheets, three frames per clip at the start, middle and end, and looks at them to confirm what each shot shows. The sheets appear under its steps in the chat so you can look too.
- Picks each moment. From each clip it takes the steadiest stretch the length of the shot, then checks the exact frames there: the face in frame, the plate landing. Shaky clips are swapped out or flagged.
- Reads the music. It finds the tempo, every beat, and where the groove kicks in after any soft intro; the reel's audio starts on that beat.
- Builds the cut. Every shot gets a length in beats (two for quick inserts, three or four for room and DJ shots, five or six for the opener and closer), the beats become exact frames, and the shots land on V1 with the music on A1. Every cut is on a beat, and the lengths add up exactly to the end.
- Checks its work. It pulls one frame from the middle of every shot into a strip, looks for people half out of frame, duplicate shots, a wrong moment, and fixes what it finds.
- Hands back a version. It saves the cut as a version with a description, reports the shot list with timestamps, and renders when you ask.
Giving notes
Talk in times and clip names: "remove the shot at 13 seconds", "replace the crate shot at 4 s with DJ_0147 from 16 s on", "make the opener longer". The agent edits its shot list, rebuilds on the same grid, checks the strip again, and saves the next version. Switching to a different song keeps the shot list; only the beat grid changes.
Everything it does is an undo step in the Edit view, and every turn is also a minor version in the Versions menu, so nothing is lost.
Where the pictures go
Contact sheets, frame grabs and QC strips are image assets under video/analysis/ in the project. Beat grids are small JSON text assets next to them. Shake analyses are cached per file, so re-cutting the same footage is fast.
Limits in this version
- No fades, slow motion or split screens yet; the agent cuts straight.
- Vertical crops of horizontal footage are a transform per clip (scale and offset), which you can adjust in the Clip panel.
- The agent needs the app window open: titles are drawn by the app, and the chat provider needs an API key under Settings.