AI video editing prompts for footage you already recorded

Five copyable briefs for cuts, captions, reframing and revisions, with a check for each result.

Om Patel

Written by , Founder, Rushilo

Updated · 6 min read

We make Rushilo.

A MacBook Pro screen filled with lines of code in an editor
Photo by Caspar Camille Rubin on Unsplash

The short answer

Short answer

To prompt an AI video editor, identify the source clip or selected timeline range, describe one concrete change, state what must stay, and define how you will check the result. Ask for a proposed edit before a large change. These prompts are for cutting and arranging existing recordings with an agent connected to an editor; they do not generate replacement footage or guarantee a good edit.

Start with an editing brief

“Make this better” leaves the agent to invent your priorities. A useful brief tells it which recording matters, who will watch, and what the viewer should understand. Give the actual filenames and the selected range. If the transcript contains an uncertain name or number, supply the approved spelling or flag it for review.

Use this reusable structure. Replace the example values with your project details; the duration is a target, not permission to discard necessary information.

Source: selected intro in demo-take-2.mov.
Audience: a new user trying the export feature.
Goal: show where to export, then show the resulting file.
Change: propose cuts to shorten the intro to about 20 seconds.
Preserve: the export limit, the warning and the complete button click.
Review: list source ranges and reasons before applying cuts.
Delivery: keep the project editable; do not upload or publish.

The same brief can guide a human editor. An agent also needs tools that can carry it out. Rushilo exposes transcript, timeline, checkpoint and export tools to a connected agent. Clueso also documents editable videos through MCP. A prompt cannot add a capability that the connected editor lacks. Check the available tools before requesting a specialized effect or automatic tracking.

Five prompts for common editing jobs

These are suggested instructions, not transcripts from tested sessions. Start with a short duplicate project and adapt them to the operations your editor supports.

1. Shorten an introduction

Read the selected intro and its timed transcript.
Propose a shorter opening that starts with the viewer's task.
Keep the sentence explaining who this feature is for.
List the exact source ranges to remove and the resulting wording.
Do not cut yet. Flag any sentence whose meaning would change.

Approve the words before the agent changes the timeline. A short opening is not automatically a useful opening: the viewer still needs enough context to understand the demonstration. For repeated attempts at the same line, use the bad-take selection workflow rather than choosing the last attempt by default.

2. Tighten pauses without cutting the demonstration

Find long gaps between spoken words in this selected section.
Propose pauses to shorten, with source times.
Keep pauses while the cursor performs a required action,
while a result loads, or while the viewer reads an important label.
Leave breaths and sentence endings natural. Do not change speed.

Silence can contain useful picture. Listen and watch together before approving cuts. If you are using Rushilo's transcript tools, word indexes change after a cut, and the cut removes that time across tracks. Ask the agent to reread the current transcript before its next group of cuts. See silence and filler-word removal for the separate cleanup controls.

3. Prepare captions for review

Transcribe this finished dialogue edit with the local speech model.
If the local model is unavailable, stop and tell me; do not switch
to a paid cloud provider without asking.
Use this spelling list: [insert names and product terms].
Flag uncertain numbers and negations. Keep the spoken meaning.
Make captions for review, then show what needs correction.

Check the audio yourself, particularly names, amounts and words such as “not.” A vocabulary list helps communicate intent but is not evidence that the transcription is correct. Decide whether the delivery needs burned-in or closed captions before styling the entire video.

4. Reframe a short excerpt

Duplicate the approved landscape edit for a vertical version.
Use only the selected excerpt. Keep the demonstrated control
and its result visible, even if that means using a wider layout.
Propose framing at the start, middle and end before changing it.
Do not assume face tracking or recreate anything outside the frame.

Cropping can hide the very step the narration describes. Review the action, not only the speaker's face. The landscape-to-vertical guide covers the visual tradeoffs. If the source is too wide for a readable crop, use a new layout or record a closer view.

5. Make one precise revision

In the current timeline, change only the title over the export shot
from "Unlimited exports" to "100 exports per month".
Keep its timing, position and style. Do not alter the dialogue.
Show the changed frame and report exactly what changed.
If that text is baked into the recording, stop and explain options.

Name the current object and desired value. “Fix that bit” can refer to the wrong version after an earlier edit. If you point to a time, say whether it is a source time or the current timeline time. A reread of the project prevents an old cut list from silently becoming a new instruction.

Check the result with an acceptance sheet

Save a named version before a large change. Review one job at a time: dialogue first, framing next, captions after the cut settles. A single instruction to cut, restyle, translate and publish creates several things to investigate when the result is wrong.

Dialogue cut

Accept when
The intended statement remains complete
Repair when
A condition, correction or useful action disappears

Captions

Accept when
Words match the final audio and important sounds
Repair when
Text is guessed, mistimed or covers the demonstration

Reframe

Accept when
The required action stays readable
Repair when
The crop hides controls or results

Revision

Accept when
Only the named object changes
Repair when
Other tracks or earlier decisions move

Keep a record with the prompt review sheet. Record the request, what changed, the check and any repair. This measures useful output on your project; counting tool calls or seconds deleted does not tell you whether the edit improved.

When a prompt does not work

First ask the agent to describe the current selection and available operation. If it has the wrong clip, fix the reference. If it lacks the operation, use a supported method or change tools. If it misunderstood the editorial goal, supply the intended final wording or a concrete frame example. Repeating “make it professional” rarely resolves which of these failed.

In Rushilo, connect your agent, select a short excerpt, and start with a proposed cut list. Review the exported excerpt outside the editor before applying the method to the full recording. Your agent account or API usage is separate from the editor licence; use the cost worksheet when evaluating that setup.

Frequently asked questions

Can I use the same video editing prompt with Claude Code and Codex?

You can reuse the editorial brief, but the connected editor must support the requested operation. Identify the current clip or selection, explain the change and preservation rules, and check the output. A prompt alone does not supply a timeline, renderer or unsupported effect.

Why does an AI editor cut the wrong section?

Common causes include an ambiguous clip reference, confusing source time with timeline time, and reusing word indexes after a cut. Ask the agent to read the current selection and transcript, then propose the exact ranges before changing them.

Do these prompts generate new footage?

No. These templates describe cuts, captions, framing and revisions to existing recordings. Generating or replacing footage is a different operation and requires separate tools, permissions and review.

What is the best way to prompt an AI video editor?

Name the source clip or timeline range, describe one concrete change, state what must stay, and say how you will check the result. Ask for a proposed edit before any large change. With Rushilo, Claude Code or Codex can then apply the approved cuts on a visible timeline you can undo. The prompt and your review still decide how good the edit is.

Sources

  1. Rushilo features and connected-agent workflowChecked October 19, 2026.
  2. Rushilo release historyChecked October 19, 2026.
  3. Clueso MCP: editable video workflowsChecked October 19, 2026.

Cite this page

Last verified: Oct 19, 2026

Patel, O. (2026, October 19). AI video editing prompts for footage you already recorded. Rushilo. https://rushilo.com/blog/ai-video-editing-prompts
Om Patel

About the author

Founder, Rushilo

Om Patel is the founder of Rushilo, a desktop video editor driven by the user's own AI, and of BigIdeasDB, a research platform that reads real complaints and reviews to help founders validate products. He writes Rushilo's research articles with the same method: primary sources, dated numbers and the words real users write.

Edit video with the AI you already pay for.

Rushilo is a desktop editor for Mac and Windows that Claude Code and Codex drive on a timeline you can see. Your files stay on your computer.