A ChatGPT or Claude workflow to auto-edit talking-head videos

Short answer

Write the script and hook in ChatGPT or Claude, film several takes, then hand the footage to one editor. To steer the edit from the chat, connect Descript or OpusClip's AI Producer; both connect to Claude and ChatGPT. With a computer and Claude Code, one skill such as AI Editing Studio runs the editing steps locally.

Disclosure: I make AI Editing Studio, path (b) on this page. Everything about the other tools comes from their own product, pricing or documentation pages, linked where each tool appears, and was checked on November 11, 2026. I have not benchmarked their output against each other.

The eight steps at a glance

Each step and its tools

1. Script and hook with ChatGPT or Claude

Use the assistant you already pay for: ChatGPT or Claude. Ask for short spoken lines, one idea per line: best-take selection, dead-air cuts and captions all work line by line. A brief for either app:

Write a 40-second talking-head script for Instagram Reels about [topic] for [audience]. Give 5 hook options under 12 words, then the script in short spoken lines, one idea per line, ending with one clear call to action.

2. Film several takes

Say every line two or three times; a fluffed line costs nothing when the editor picks the best take. CapCut's teleprompter sits on the recording screen of its mobile app, scrolls while you shoot, and adjusts speed, font size and color; CapCut calls it free. BIGVU offers a teleprompter on web and mobile. Film vertical, in 4K if your phone allows, so an editor can punch in without a soft picture.

3. Transcribe and pick the best take

Descript turns the recording into text you edit like a document and says you can "cut filler words and retakes in a click"; its pricing page lists Remove Retakes as limited on the free plan and included on paid plans. Gling, built for YouTube creators, removes bad takes, silences and filler words. In a terminal, faster-whisper returns word-level timestamps on CPU or GPU, the raw material for any take-picking script you ask Claude Code to write.

4. Cut dead air

Descript and Gling both cut pauses. Submagic lists "AI Remove bad takes & silences" on its Pro plan. Captions detects pauses and filler sounds such as "um" and can remove them, with a "Restore original" undo. auto-editor is a command-line tool that cuts by audio loudness or motion and can export a Premiere Pro, DaVinci Resolve or Final Cut Pro timeline instead of a finished file.

5. Captions

Submagic is caption-first: it "adds captions, removes silences, inserts B-roll, and finds key moments automatically" for TikTok, Reels and Shorts. CapCut's auto captions are free on desktop, web and mobile and can translate. Read every caption once before export; names and jargon are where speech recognition slips.

6. Motion graphics and B-roll

• Captions AI Edit takes an unedited single-speaker clip of up to 3 minutes, a style and, on the web, a prompt. It adds "auto captions, transitions, B-roll, music, and motion graphics"; its "Generate media" option creates B-roll at a higher credit cost.
• AI Producer by OpusClip is edited by chatting, in the browser, through a Claude connector or a ChatGPT plugin. Its documentation lists removing filler words and pauses, adding captions, motion graphics, B-roll, music and sound effects, and recreating a reference video's style.
• Descript says it can generate missing B-roll, and through the Descript MCP server an assistant can add stock media and overlays on request.
• Pexels is the manual route: free stock videos, no attribution required.
• HyperFrames by HeyGen renders HTML, CSS and animations into MP4 and installs as a Claude Code plugin, for custom graphics you describe to Claude.

7. Sound

Descript's Studio Sound removes background noise and enhances the voice. Adobe Podcast Enhance Speech calls itself a "free AI filter for cleaning up spoken audio". Pixabay hosts royalty-free sound effects for cuts and graphics.

8. Export and check the safe zone

Export vertical 9:16. AI Producer exports one MP4 up to 1080p at 30 fps; Descript's MCP server publishes a share link and download URL and can export a Premiere Pro or DaVinci Resolve timeline. Then watch the file on your phone: Reels lays your caption and name over the bottom and its buttons down the right edge, so captions and key text belong in the middle.

Path (a): assemble separate tools

• Lightest: ChatGPT or Claude for the script, then CapCut for teleprompter, captions and export. Works on a phone; the most manual.
• Most automatic in one app: Captions AI Edit or Submagic after the script.
• Driven from the chat: Descript's MCP server works from Claude desktop, web and Code, from ChatGPT's app directory and from other MCP clients, at no separate charge beyond Descript AI credits. AI Producer connects to Claude, and its ChatGPT plugin runs in Codex on paid ChatGPT plans; editing steps use OpusClip credits.
• Editor in the loop: Gling or auto-editor for takes and dead air, then Premiere Pro, DaVinci Resolve or Final Cut Pro for graphics by hand.

The price of this path is the hand-offs: each tool has its own account, plan and export, and a late change can mean redoing the steps after it.

Path (b): one Claude Code skill for steps 3 to 8

AI Editing Studio by Artem Grigorets (me) takes over after you film. Free Studio is a free ZIP download with no email gate; it installs in ~/.claude/skills/studio-edit and runs with /studio-edit in Claude Code.

• Steps 3 and 4: transcribes your takes with word-level timing, keeps the best take of each line, cuts dead air and punches in from 4K.
• Steps 5 and 6: cuts you out of the background so big words sit behind your head, then adds captions, motion graphics and animated inserts built by sub-agents. For B-roll it imports the video clips and PNGs you give it.
• Step 7: adds sound and your music; Pixabay sound effects download during setup.
• Step 8: runs Instagram safe-zone and frame checks. You watch the result, say what to change in plain words, and each round re-renders in about two minutes.

Editing runs on your computer; the website does not upload raw footage. It is built on HyperFrames and faster-whisper, and you can build your own style with your fonts, colors and sounds. Requirements: Mac, Windows 10/11 or Linux; a Claude plan with Claude Code; ffmpeg; Node.js 22+; Python 3.10+; about 10 GB free disk; roughly 2 GB of models on the first Reel.

Free covers the talking-head format with 5 motion effects and 5 caption styles. Studio Pro is $49 once and lists 10 formats, 50+ motion effects and 50+ caption styles. Pro is currently a paid preorder: no announced release date, a release update by email within two weeks, and a full refund within 14 days. Claude and ChatGPT editions are being tested; the exact supported apps will be disclosed before release. More Claude Code options: Claude Code skills that edit video.

The two paths side by side

Who this is not for

• Phone-only creators. Path (b) needs a computer and a terminal; stay with CapCut or Captions.
• Long horizontal YouTube videos or multi-camera shoots. Gling is built for YouTube creators; AI Producer's web app does not support multi-camera edits yet; Studio is built around talking-head Reels.
• Client work with a frame-exact brief. Use a timeline editor; auto-editor and Descript can hand it a timeline.
• Anyone who wants no AI subscription. Step 1 assumes ChatGPT or Claude, chat-driven editors bill in credits, and Studio needs a Claude plan.

Questions

About the author

Artem Grigorets writes about making content with AI, content automation and building a personal brand with AI, and makes AI Editing Studio for creators who film talking-head Reels. Instagram: @rtgrigorets · Newsletter: grigorets.com · Support: value@grigorets.com

AI Editing Studio is a product by Elvo LLC, 30 N Gould St Ste R, Sheridan, WY 82801, USA. Claude is a trademark of Anthropic. Studio is not affiliated with or endorsed by Anthropic or OpenAI. Other names belong to their owners; each is linked to its own page above.