Caption Silent Footage With the DepCut Agent

A b-roll montage, a music-only reel or a screen recording without narration gives a transcriber nothing to hear. Run Generate subtitles on one of these and the Subtitles panel tells you that no speech was found in the cut. No captions are added, and the button changes to Try again.

That message is your cue to caption what the picture shows.

Ask the Agent for visual captions

1. Make sure the footage sits on the main video track. The Agent samples frames from that track.
2. In Subtitles, choose the caption language in the menu. The captions are written in the language set on the track.
3. Open the Agent from the top bar, or press Cmd+J or Ctrl+J, and ask for what you want. Something like Caption this video from what is on screen works.

The Agent captures small frames spread across the cut, each stamped with its time in the video. An AI model looks at them and writes short narration captions for what is happening, in order, timed to the cut.

What it costs

Captioning from visuals is AI work. On a project stored on your Mac it runs through your own Claude login. On a cloud project it runs on DepCut's hosted AI and spends account credits.

Edit the result like any other caption

The new lines are ordinary captions on the track. Fix the wording in the Content tab, drag their bars on the timeline to change when they appear, and restyle them in the Styles tab.

A few things to know:

• The captions describe the footage. They are a narration written for the viewer, and nobody said them.
• The model sees small frames, so check names, products and places. Correct anything it guessed wrong.
• The lines carry no word timings. If you switch on Highlight spoken word, DepCut spreads the words across each caption in proportion to their length.

Get better captions from your footage

• Footage with clear actions and scene changes gives the model more to describe than one long static shot.
• The Agent samples a limited number of frames from the whole cut, so shorter videos get the most detail per caption.
• Over busy visuals, choose a plate style such as Clean or Block so the text stays readable.
• Keep the first caption doing the work of a hook. If it reads flat, edit it by hand in the Content tab.

Turn the captions into a voiceover

Once the lines read well, you can ask the Agent to read the captions aloud. DepCut places the generated voice on the soundtrack and re-times the captions to match its pace. Voice generation spends account credits.

This gives a silent clip a spoken narration without writing a script from scratch. Play it through once to confirm the voice and the picture agree.

Caption your silent clip

Drop the clip into DepCut, open the Agent and ask for captions from what is on screen.