Regenerate video (beta) is an optional step you can take when updating audio with Descript’s regenerate tool. If you've changed the script or smoothed out a rough edit, video regenerate updates the speaker’s visual performance to match the audio—keeping lips and speech in sync. It’s especially useful for fixing mistakes or improving delivery in on-camera recordings without having to re-record.
This article covers how to use Video Regenerate to sync visuals with updated audio in your project.
To learn more about editing the audio itself, see these related workflows:
- Use Regenerate to edit or change recorded audio
- Use Regenerate to fix or smooth out any jumpy, abrupt, or awkward audio
If you’ve changed the script, you’ll also need a custom AI speaker with consent and speech generation enabled to use video regenerate. Here’s how to create and assign an AI speaker
How to use video regenerate
- In your script, highlight the section you'd like to fix (a choppy transition, abrupt cut, or mistaken word). Including a few additional words around the section can help with matching the tone and pacing naturally.
- Click Regenerate from the hover toolbar that appears.
- Enable the Regenerate video toggle, then review the consent statement. To continue, check the box confirming you agree to Descript’s use of your visual performance. This step is required to use Video Regenerate.
- Click Regenerate to generate both the new audio and synced video.
Descript will first regenerate the audio. Once that's complete, it will begin regenerating the video. You'll see a green checkmark when the update is finished. This can take several minutes.
Video regenerate tips and best practices
- This feature is in beta, and still being improved. Learn more about beta labels in Descript.
- Video regenerate won't work on B-roll. It can only be applied to media in the script track
- Avoid regenerating video over existing edit boundaries. Regenerate works best on continuous, uncut sections of video.
- Single speaker only. Multi-speaker video is not supported
- Speaker should face the camera. Side angles reduce quality
- Face should fill a reasonable portion of the frame. Not too close, not too far
- Mouth must be clearly visible. No hands, mics, or objects in the way
- Up to 4K video is supported — but results can vary! Regenerate tends to work most reliably with non-4k video.
How it works
Video Regenerate happens in sequence: first, Descript regenerates the audio. Then, it uses the updated audio to sync your speaker's visual performance.
If your script has changed, Descript uses your assigned AI speaker and avatar to generate both the voice and the synced video. If you're just smoothing the audio or healing pacing, visuals can still be regenerated—but don’t require consent for speech generation.
Video generation may take a few minutes depending on the length of your clip.
When you enable the Regenerate video toggle, Descript will ask you to confirm consent for modifying the speaker’s face. This is required before visual regeneration begins and ensures your likeness is used securely.
You’ll need to check a box acknowledging that the process uses data that may be considered biometric. This step appears every time you regenerate visuals.
Limitations
- Only works on A-roll (camera/script track)
- Not supported in sequences
- Up to 250 characters per regenerate action
- Single speaker only; multi-speaker clips aren't supported
- Only works in English
- Stock voices are not supported—must use a custom AI speaker
You won’t be able to regenerate if:
- The selection includes multiple edit points
- The selected text crosses a scene boundary
- The selected text is ignored
- You edited the script but don’t have access to the speaker’s AI voice
- You’re trying to regenerate a clip that’s already been regenerated — click Keep to convert it first
Learn how to replace audio with Regenerate or smooth edits and pacing using Regenerate.