AI Video Editing in Claude Code: A Guide for Non-Editors

The author of the video admits she can't edit video at all — yet the whole video was edited by AI: from raw footage with pauses and failed takes to a finished video with animation, captions and reels. No complex prompts — the task is explained to the AI in plain words. We broke the process down step by step and added our notes and short prompt templates.

Watch the video on YouTubeThe breakdown this guide is based on (in Russian). The author's full prompts are in the video description.

How it works

Claude doesn't "watch" video like a person. First the recording becomes a transcript with a timecode for every word. Claude reads it like a script and decides where the pauses, slips and key points are. Free tools that Claude installs and runs itself do the cutting. You talk to it like to an editor: explain what you want and give feedback. The main idea: the AI is about 10% of the result, the other 90% is how you set the task.

What you need

Setup and brand book

Put your footage in a folder and ask Claude to install everything needed. Then build a brand book from references you like — otherwise you get the default "AI look".

The folder [path] contains my video footage. Install everything needed for editing: ffmpeg, the Whisper speech model, word-level audio alignment and background removal. Check that every file opens and tell me its duration, resolution and frame rate. Do not edit anything yet.

The references folder has examples I like. Study them and create a brand book for my videos in brandbook.md: mood, 2–3 HEX colors, free fonts, plates and highlights, reel caption style, motion rules and what we never do. Show it on a frame from my video and do not edit until I approve.

Steps 1–2. Transcript and rough cut

The transcript must include everything you hear — fillers, slips, repeats and pauses. Then ask for a list of cuts first and cut only after your approval. Shorten pauses to about 0.2 s (never to zero), keep fillers inside continuous phrases, keep the last good take and start exactly on the first word.

Make a rough cut from the transcript. First show me the list of cuts with timecodes and reasons, and do not cut until I say "cut". Shorten pauses over 0.4 s to 0.2 s, remove fillers, off-topic talk and failed takes (keep the last good one). Do not clip the start or end of words. Start on the first word.

Steps 3–5. Chapters, key points and animation

Add short chapter titles in the corner, put terms, numbers and lists on screen in sync with speech (no more than 15 seconds without graphics), and make the strongest motion in the first ~30 seconds — the hook. Always ask for a list first, then approve.

Steps 6–8. DaVinci Resolve, reels and captions

Connect Claude with DaVinci Resolve and ask it to lay out the project on separate tracks so you can fix small things by hand. Then ask for 5 reel ideas with hooks, edit them in 9:16 in the same style, and add word-by-word captions that stay below the face and outside the social media safe zones.

How long it takes

The author's first video took about 3.5 hours of Claude's work plus a couple of hours to figure things out; the second took about 3 hours. A reel takes about half an hour. Claude works in the background, and a month of the plan costs about as much as an editor charges for one reel.

This guide is based on the video "This video was edited by AI" (in Russian). The author's full prompts for every step are in the video description.