Claude Code video editing: how AI coding agents edit video from one prompt

Short answer

An AI coding agent edits video by writing and running code: it reads a transcript, decides where to cut, calls command-line tools or rendering frameworks, checks the result and repeats. With the right template and tools in place, one prompt can produce a finished, captioned video. Without them, the agent has nothing to work with.

• For cutting raw footage into a finished video: video-use, an open-source Claude Code skill that removes filler words and dead space, grades color and burns subtitles.
• For motion graphics and designed videos: Remotion, which renders video from React code and publishes official Agent Skills for Claude Code, Codex and others.
• For removing pauses and mistakes: auto-editor, a free command-line cutter with its own agent skill.
• For cloning a proven video format at volume: Hypit, an open-source runtime built for agents.
• For clipping long videos into shorts: an AI clipper’s MCP server, such as OpusClip’s or Vugola’s, so the agent can run clipping jobs instead of rebuilding them.

The one-prompt edit, step by step

On October 18, 2026, creator Riley Brown posted a 97-second video on X with the line: “This video was edited with 1 prompt with Opus 5.5.” He wrote that he uploaded a single video of himself “with many mistakes”, and listed 15 things the agent did. In his words and order, summarised:

1. Analyzed the transcript.
2. Searched Twitter (X) for relevant B-roll; he adds that “Claude Code literally calls GrokBot for this step.”
3. Scraped the video assets from those posts.
4. Cut the video based on mistakes and word gaps.
5. Built a Blender scene from a template.
6. Built the interface and components for each of the scene’s four shots: a computer, a phone, a table and a board.
7. Took screenshots of every section, analyzed them and fixed problems.
8. Synced every shot, zoom and animation to the exact words spoken.
9. Animated a robot arm placing real logos on the table as each was mentioned.
10. Drew a whiteboard ending with logos and graphics instead of text.
11. Pulled a cutaway from another creator’s video, with his voice.
12. Rendered all 3,487 frames on RunPod cloud GPUs, which he says took under 20 minutes and cost less than $1.
13. Showed him in the bottom third only while he was talking, keeping B-roll full screen.
14. Added word-by-word captions, sound effects and music.
15. Uploaded the result to his Google Drive, ready to post.

Two details in the same post matter for anyone hoping to copy it. The Blender scene came from a template he had built, and he said his planned YouTube explainer would not be “a copyable skill” but would teach the foundations so people can build their own. In other words, the single prompt worked because a lot of preparation sat underneath it. The render time and cost figures are his own report; we have not reproduced them.

How agent video editing works

Strip away the specifics and the pattern is the same in every agent workflow:

1. Understand the footage. The agent transcribes the video with word-level timestamps, so every cut can be anchored to a word rather than a guessed second.
2. Decide the edit. It reads the transcript to find mistakes, pauses, repeated takes and the moments worth keeping, then writes an edit list.
3. Execute with tools. It runs tools that do the actual work: a cutter for silence, a rendering framework for graphics and captions, a 3D program, or a clipping service through its API or MCP server.
4. Check its own output. Like Riley Brown’s step 7, a good workflow renders stills or previews and has the agent inspect them before the final render.
5. Render and deliver. Rendering can happen locally or on rented GPUs, and the file is saved or uploaded.

The agent is the director, not the camera. Its quality depends on the tools it can call and the template it starts from.

Tools agents use, compared

video-use: raw footage to final.mp4

video-use, from the team behind Browser Use, describes itself as a way to “edit videos with Claude Code”, 100% open source under the MIT licence. You drop raw footage in a folder, start Claude Code (or Codex, or another agent with shell access) there and ask for an edit; according to its README, the agent inventories the sources, proposes a strategy, waits for your approval and then writes edit/final.mp4 next to your files.

The README lists cutting filler words and dead space between takes, automatic color grading, 30 ms audio fades at every cut, burned-in subtitles, and animation overlays generated through HyperFrames, Remotion, Manim or PIL in parallel sub-agents. It also says the agent checks the rendered output at every cut before showing you anything, and keeps session notes so the next edit picks up where the last one stopped. The model never watches the video; it reads a word-level transcript from ElevenLabs’ Scribe, so you need an ElevenLabs API key and ffmpeg. A hosted version has a waitlist. Best for: talking-head, tutorial and interview footage where most of the work is cutting.

Remotion: video as code

Remotion is a framework for making videos with React. Its documentation has a guide titled “Prompting videos with coding agents” that names Claude Code, Codex, Kimi Code and OpenCode, and walks through creating a blank project, installing the Remotion Agent Skills and starting the agent in the project folder. The skills, which Remotion says suit agents such as Claude Code, Codex, Kimi Code and Cursor, include a best-practices skill and commands to create projects and compositions.

Remotion is the natural choice when the output is designed rather than cut: animated explainers, captioned talking heads with graphics, data-driven videos or templates you re-render with new words. Check the licence before commercial use. Remotion’s pricing page says its Free License covers individuals and companies of up to 3 people, and a paid Company License is required for collaborations and companies of 4 or more.

auto-editor: the first pass

auto-editor removes “dead space”, by default the quiet parts of a recording, and can also cut by motion, pad each cut, speed up sections instead of removing them and export a timeline to Premiere Pro, DaVinci Resolve, Final Cut Pro, Shotcut or Kdenlive. Its README offers an agent skill install, which lets an agent run the first pass of Riley Brown’s step 4, cutting pauses, without custom code. It is free and open source, which also makes it useful outside agent workflows; we list it among free Opus Clip alternatives for that reason.

Hypit: cloning a format at scale

Hypit’s home page describes it as a way to “clone any viral video with AI agents”: you give your agent a reference video, and Hypit rebuilds its shots, pacing, captions and effects with your content, anchoring edits to words instead of seconds. It installs as an agent skill, works with Claude, Codex or any agent according to the site, and lists an open-source core with $0 software cost, with models either brought by you or used through Hypit’s API. It suits marketers producing many variants of a format that already works, rather than one-off creative edits.

Opus Clip and Vugola MCP servers: clipping on command

If your goal is the usual clipping job, long video in and short vertical clips out, it is usually better for an agent to call a clipping service than to rebuild one. Two of the tools we track publish MCP servers for this.

OpusClip runs a hosted MCP server that works with Cursor, Cline, Claude Code, ChatGPT/Codex and other MCP hosts, with OAuth sign-in and no local install. Its documentation is clear on access: any account can sign in and see the tools, but calling one requires Pro Beta, Max or Business plan access. Our Opus Clip API guide covers the plan rules and limits.

Vugola documents an official OAuth MCP server that lets agents such as Claude Desktop, Claude Code, Cursor, Cline and Windsurf clip videos, download clips to disk and manage scheduled posts and channel Automations, alongside a REST API. See our Vugola review for the plans and documentation conflicts we found.

Agent or AI clipper?

The two approaches combine well: an agent can call an AI clipper for the clip picks and then use code to brand, caption or assemble them. For a broad view of dedicated tools, see the best AI clip makers guide.

What it costs

• The agent. Remotion’s guide notes that most coding agents require a paid subscription.
• Rendering. Short renders run on a laptop; longer or 3D work may need rented GPUs, as in the example above.
• Licences and plans. Remotion is free only for individuals and companies of up to 3 people, and the OpusClip MCP needs a qualifying paid plan.
• Model calls. Transcription, image or voice generation through paid APIs adds per-video costs.
• Setup time. Templates and tools take hours to prepare; the payoff comes when you reuse them.

For clipping specifically, compare this against a subscription using our AI clipping cost-per-minute table.

Footage rights and review

Agent workflows make it easy to pull in other people’s material. In the example above, the agent scraped B-roll from posts on X and pulled a cutaway from another creator’s video. Before you automate that, make sure you have the right to use the footage, logos and voices involved, and check each platform’s terms. Keep a human review step before anything is published: agents can misread a transcript, cut a word mid-sentence or place a logo on the wrong line, and you are responsible for what goes out under your name.

How to try it

1. Install a coding agent such as Claude Code or Codex, plus Node.js.
2. Follow Remotion’s guide: create a blank project with npx create-video --yes --blank my-video, then install the skills with npx remotion skills add and start the preview with npm run dev.
3. Start your agent in the project folder and describe a short video, for example a 20-second captioned clip from a transcript you provide.
4. For real footage rather than motion graphics, follow video-use’s README instead: it has a setup prompt you paste into Claude Code, and it asks for an ElevenLabs API key. Add auto-editor if you only want pauses cut.
5. Review every render before you post it, and keep the prompt and template so the next video takes one prompt too.

When we reproduce an agent workflow ourselves, we will log it with the same fields as our clipping tests.

Frequently asked questions

Sources

Vendor facts on this page come from these public pages, checked October 19, 2026. Prices and plan limits change; confirm them at checkout.

• Riley Brown on X: “This video was edited with 1 prompt”
• video-use on GitHub
• Remotion: Prompting videos with coding agents
• Remotion: Agent Skills
• Remotion: License & Pricing
• auto-editor on GitHub
• Hypit home page
• OpusClip MCP Server documentation
• Vugola API and MCP documentation