Video editing guide

AI Video Edit Prompts That Produce the Exact Edit

The assistant can only make edits the editor supports, so the best request is one that names a clear operation, the place it happens and anything that must not move. Here is how to write that kind of request and check the result.

Put the action, target, time and limits in one sentence

Begin with the verb: cut, add, mute, lower, speed up, resize or zoom. Then say what it applies to, the selected clip or a range such as 00:40 to 00:44. Finish with what should stay the same, like the audio or the order of clips. That sentence maps directly onto editing commands.

Use the times shown on the timeline now. Every removal moves the material after it earlier, so read the playhead after each cut instead of working from a list you wrote while watching the original.

Cut 00:40 to 00:44 and close the gap. Leave the volume and playback speed as they are.

Quote the exact words for any title

Put the text in quotation marks, then give the position and the start and end time. “At the beginning” is vague; “from 00:00 to 00:03 at the bottom” is not. Short titles on a dark background stay readable over changing footage.

Watch the title play, not just the first frame. Footage that is dark at the start may turn bright a second later, and a title that fits a landscape frame may wrap after you switch to vertical.

Add “Day 1 in Lisbon” at the top from 00:00 to 00:03 in white with a dark background. Do not change the clips.

Ask for cuts with exact boundaries

Give a start and end for every section you want gone, and check each join. A cut in the right place can still clip the first sound of a word or leave a jump in someone’s position.

For pauses, the editor can measure quiet ranges on your device and pass them to the assistant. It does not know the meaning of the words, so give your own times for anything important and listen to every join.

Remove 01:05 to 01:09 and join the footage either side. Keep the speed unchanged.

Request zooms and punch-ins precisely

Say whether you want a fixed zoom on the whole clip or zooms over particular times. For timed zooms, say whether each should snap in instantly as a punch-in or move gradually, and give the percentage. The maximum is 400%.

A punch-in keeps the sound and timing; it only changes the framing for those seconds. Check that faces and important details stay inside the frame at the larger size.

On the selected clip, punch in to 125% from 00:06 to 00:09 and from 00:14 to 00:17. Instant cuts, no fades.

Pick a model for the request, not for the effect

Quick and Standard use OpenAI language models and Advanced uses an Anthropic model. They all issue the same set of editing commands, so a more expensive model cannot add object removal, automatic captions or generated footage.

Select a clip when the request is about that clip, and keep to one or two related changes per request. The estimate shows the most credits the request can reserve.

Check each edit before stacking the next one

Play the changed section with sound and compare it with the previous step using undo if you are unsure. If the edit touched the wrong clip, undo first and repeat the request with the right clip selected, rather than piling a correction on top.

Small timing fixes are faster by hand and cost nothing. Keep the original video and download a project backup so you can rebuild the edit later if browser storage is cleared.

Frequently asked questions

Does a longer prompt give a better edit?

Not on its own. A precise action, target, time and limit matters more than adjectives. Split unrelated changes into separate requests.

Why did my second request change the wrong moment?

Your first edit probably shortened the timeline. Use the times shown now, not the ones from the original recording.

Can wording unlock an effect the editor lacks?

No. The assistant can only issue the commands the editor supports, whichever model you choose.