AI Talking-Head Editing

AI talking-head editing turns your recording into a transcript, then asks AI to find what can be cut or should be corrected. By default, AI only suggests changes; nothing on the timeline changes until you approve. In the interface, it lives in the Smart Clipping workspace on the left of the editor, under the Transcript Clipping task.

This feature is still being improved. AI suggestions can be wrong, so review each one before applying.

1. Prepare: a transcript and an AI model

Transcript: the first time, choose a local transcription model and click Download & Transcribe. Transcription runs entirely on this Mac and audio never leaves it. You can also import an existing SRT or VTT file.
AI model: in Settings > AI & Transcription, configure a model behind an OpenAI-compatible endpoint (service URL, model name and API key) and click Verify & Save. Without one you can still browse and hand-edit the transcript, but AI suggestions are unavailable.

2. Generate edit suggestions

Under Clean to a Reference Draft, choose what to do, then click Generate Edit Suggestions:

Option	What it does
Correct transcript	Fixes clear recognition, terminology and punctuation errors without cutting media
Remove filler words	Marks unnecessary fillers such as "um", "uh" and "and then"
Remove retakes	Finds false starts, self-corrections and adjacent duplicate takes
Shorten word gaps	Uses local audio analysis to locate long gaps; choose Compact, Natural, Relaxed, or keep all pauses
Edit for clarity	Low-intensity suggestions for clearly rambling or off-topic content; these are subjective

To steer the result toward a script you already wrote, fill in the Reference Draft and AI will optimize toward it. The reference draft is saved only in the current project.

3. Review each suggestion

Filter the list by All, Cuts or Text fixes. Every suggestion can be revealed on the timeline and played:

For cuts: choose Remove from video or Keep.
For text fixes: choose Use Suggestion or keep the current text.

The footer shows how many are selected and about how much shorter the video will be.

4. Apply to the timeline

When you confirm, everything is applied at once as a single edit, which you can take back with Undo Changes or ⌘Z. The original audio is never modified.

5. Manual cuts and the glossary

You can also work on the transcript without AI: click a word, drag to select a phrase, then choose Remove from Video, Edit Text, or play that part of the video.

When AI keeps misspelling a product name or technical term, select it and choose Add to Terminology…, then enter the preferred spelling and a description. A term can apply to this project only, or to all projects from the Glossary in Settings > AI & Transcription.

Privacy: your API key stays on this Mac. AI receives transcript text and word timings, never raw video or audio.