AI edit suggestions

After a recording is transcribed, the AI reads the transcript and suggests an edit: which parts to remove (mistakes, repeated takes, repeated points, filler words like "um", long pauses), up to three strong opening hooks, and the best moments for Shorts (see shorts for making them). It happens automatically; the card shows AI editing…, then the time it would save (for example AI edit · −2:15). Nothing is applied until you approve it.

The AI decides *what* to remove; the timing comes from the audio itself. Pauses are measured from the sound, and every cut lands between words, never mid-word. Each suggested cut comes with its reason, and you can switch any cut off.

⋯ → Suggest edits with AI (or Suggest edits again) lets you pick a style:

• Natural: only mistakes and repeated takes.
• Balanced (default): also repeated points, filler words and long pauses.
• Aggressive: tighter pacing.

There are limits so the AI can't remove too much (about 15% / 30% / 45% of the speech for the three styles).

Made a mistake while recording? Say "retake" and start the sentence again. Davnai removes the unfinished sentence (or the sentence before) together with the word "retake". You can undo this on the review screen.

No. It's an AI re-run: your plan includes a number of these per month for all your recordings together (Find more Shorts uses the same allowance). When they're used up, the button tells you and when they reset. See Settings → Usage and storage. Reviewing, saving versions, approving and making videos never count either. See video-minutes.

Suggestions need a transcript. If the recording has no speech there's nothing to suggest. If the card shows a failure, use ⋯ → Suggest edits with AI to try again.