How to Edit Video with ChatGPT: A Step-by-Step Guide

Yes. ChatGPT and Codex can edit video when connected to a video editor such as ChatCut on a Mac (macOS 13+) or Windows PC (Windows 10+ x64). In OpenAI’s ChatGPT desktop app, ChatGPT or Codex interprets your brief and calls tools, while ChatCut’s AI video editor applies the cuts, captions, B-roll, Motion Graphics, and audio on an editable timeline. Without a connected timeline editor, ChatGPT can plan cuts from a transcript or run code-based renderers, but it does not leave you a timeline project to adjust by hand.

ChatCut can also create media that was never recorded. Its AI video generator (available on paid plans) produces new footage from prompts and supported references, while its image, voice, music, caption, and Motion Graphics tools create supporting assets. When you edit through Codex or the ChatGPT plugin, your existing ChatGPT plan covers the agent’s work and core timeline editing uses no ChatCut credits.

This tutorial takes you from desktop setup to a finished talking-head edit, with prompts you can adapt for other formats. If you would rather see the result before the method, we tested this workflow on a real project and published what came out of it.

Watch a recorded editing session

Our September 17, 2026 tutorial follows a presenter briefing ChatGPT with recorded footage and separate microphone audio, reviewing a first cut, requesting revisions and B-roll, and opening ChatCut’s export controls. Jump to the stage you want to inspect:

• 1:48: source footage and editing brief
• 5:05: review and revise the cut
• 8:24: export and sharing options

The video shows the interface used for that session. Follow the current setup steps below. Its edited running time does not measure how long the complete workflow takes.

Quick start: edit a video in five steps

1. Install and connect ChatCut. In the ChatGPT desktop app, select Codex and send Read ChatCut.io/desktop to install ChatCut to install ChatCut Desktop and register its local MCP tools. To continue an existing ChatCut Web project instead, use the hosted Web plugin or copy the current setup prompt from the ChatCut plugin for ChatGPT.
2. Add your footage. Import your source files through My Assets in ChatCut (or attach local files in the conversation and ask the agent to import them), then state what you want to make.
3. Ask for a rough cut. Define what to remove, what must stay, the target length, and the intended audience.
4. Add the finishing layers. Request captions, B-roll, Motion Graphics, music, voiceover, sound effects, or generated media only where they support the story.
5. Review and export. Play the editable timeline, correct details through the agent or directly in ChatCut, then export the finished file.

The most useful prompt is a clear editor brief: source, audience, deliverable, required moments, and constraints.

What you need before you start

• A supported computer and desktop app. ChatCut Desktop runs on macOS 13 or newer (Apple silicon or Intel) and Windows 10 or newer (x64); Linux is not supported (ChatCut Product Facts). OpenAI’s ChatGPT desktop app lets you choose between ChatGPT (Chat or Work) and Codex. Neither the Desktop installer nor the Web plugin can be installed from the chatgpt.com website, an isolated remote workspace, or the ChatCut web editor.
• A ChatCut account. Sign in inside ChatCut Desktop (or complete browser authorization for the Web plugin). ChatCut’s Free plan includes the core editor, uploads, transcription, and standard caption editing without spending credits.
• Medium reasoning effort. Medium is enough to start and finish most editing tasks without unnecessary waiting. Raise it later for a complex multi-timeline job.
• Source footage you have permission to use. Clear dialogue helps when the edit depends on a transcript. You can also bring audio, images, logos, and other project media.
• A defined output. Name the audience, target platform, orientation, approximate length, and purpose of the video.
• Optional visual references. Brand colors, fonts, logo files, caption references, or a saved Design Style can make the first pass more consistent.

Give the agent enough direction to produce a useful draft, then refine one issue at a time.

How to edit video with Codex and ChatCut

In the current ChatGPT desktop app, Codex is an agent option alongside Chat and ChatGPT Work. Both ChatGPT Work and Codex can plan multi-step tasks and run connected tools: Work is tuned for documents and research deliverables, while Codex exposes developer views, local file access, and technical execution details (OpenAI Using ChatGPT guide). Neither mode supplies a video timeline by itself. Some creators pair Codex with code-rendering pipelines, such as Kevin Stratvert’s 16-clip ChatGPT Codex tutorial using HyperFrames or open-source shell skills like video-use that drive Remotion, Manim, and FFmpeg, to render a flat final.mp4. Connecting ChatCut instead gives Codex a multi-track timeline you can inspect and trim by hand.

For new setups on macOS 13+ (Apple silicon or Intel) or Windows 10+ x64 (Linux is not supported), install and sign into OpenAI’s ChatGPT desktop app, select Codex from the top-left product dropdown (or open the ChatCut plugin for Codex), and send:

Codex reads the Desktop installation guide and runs the official installer without needing Node.js or npm. Complete sign-in inside ChatCut Desktop, verify that chatcut_desktop tools are available, and start a fresh conversation if you just installed (agents load MCP tools when a session begins). Connecting ChatCut Desktop removes the hosted Web plugin from Codex so the agent targets local Desktop projects, including selected .chatcutproject packages while the Desktop window is closed (Edit a Desktop Project from Your Agent).

If a compatible local Codex installation is detected, Codex CLI also appears inside ChatCut Desktop under Agent Selection at the bottom of the AI composer, using your own ChatGPT account and quota (and offering GPT-6.1 Sol under Settings → Model on eligible accounts in Desktop 0.6.6+). With Codex or Claude Code (plugin or MCP), the agent’s work (cutting, reordering, removing pauses and fillers, captions, B-roll placement, and motion graphics the agent writes) is covered by your existing ChatGPT or Claude subscription and costs no ChatCut credits (Credits policy). ChatCut’s Free plan includes manual timeline editing, uploads, transcription, standard caption editing, and MP4/WebM, MP3, subtitle, or XML export without spending credits (with a cumulative 60-minute Free web/cloud export quota, while Desktop renders video locally on your computer). ChatCut credits apply only when you use the built-in ChatCut Agent or call ChatCut-hosted AI video (paid plan required), image, music, sound-effect, or voiceover generation. Import media through My Assets, run the staged prompts below, and review every cut before exporting.

A-Roll and B-Roll: the two stages of every edit

Two terms explain why the steps below are ordered the way they are.

A-Roll is the spine of the video: the footage that carries the story, usually the person speaking. B-Roll is everything layered on top to make that story clear and complete: a cutaway shot, a screen recording, a photograph, a chart, a Motion Graphic.

Almost any edit reduces to the same two stages. First get the A-Roll to say what it needs to say, in the right order, at the right length. Then add B-Roll where the spoken line alone is not enough.

This matters when you are directing an agent. Asking for cuts, captions, graphics, music, and generated footage in one prompt produces a draft where you cannot tell which decision went wrong. Finish the A-Roll first, confirm it plays, and only then start adding.

How to edit video with ChatGPT using ChatCut

Steps 1 to 5 build the A-Roll. Steps 6 to 9 add the B-Roll and deliver the file.

1. Install ChatCut in the desktop app

New installations connect your desktop agent to ChatCut Desktop on macOS 13+ or Windows 10+ x64. Start a task in the ChatGPT desktop app (with Codex selected) and send Read ChatCut.io/desktop to install ChatCut. Older links to chatcut.io/chatgpt now serve this Desktop installer, which opens the native ChatCut Desktop app and registers its chatcut_desktop local MCP connection rather than opening an embedded Browser panel. Sign in inside ChatCut Desktop, verify tool access, and open a fresh agent task so the session loads ChatCut’s tools.

If you are continuing an existing project in ChatCut Web, or want the hosted editor open inside the desktop app’s Browser panel, use the hosted Web plugin instead. You can search for ChatCut in the desktop app’s plugin store and click Install, or start a desktop task and send the Web plugin setup prompt below:

Complete browser authorization when prompted, then open a new task (or type @chatcut to wake the hosted plugin) so the session starts with ChatCut’s tools loaded.

2. Upload the footage and give ChatGPT a brief

In ChatCut Desktop, open My Assets → Upload → Files… (or Folder…), drag local files straight from Finder or File Explorer into My Assets, or ask the agent to import a readable local file path. In the Web plugin workflow, attach a video to the task and explicitly ask ChatGPT to import it into the project, or upload it through My Assets → Upload. A message attachment gives the agent context; it does not automatically place that file on the timeline.

A useful brief answers five questions:

1. What source files or takes should ChatGPT use?
2. Who is the audience?
3. Where will the video be published?
4. What must stay, and what may go?
5. What is the intended viewer outcome?

For example: “Turn this 12-minute product walkthrough into a concise YouTube tutorial for first-time users. Keep the setup and final result, remove repeated explanations and dead air, and do not add generated footage.”

3. Create the rough cut

Make the A-Roll work before adding any visual polish. Ask ChatGPT to remove verbal clutter while protecting the speaker’s meaning and natural cadence.

Open the transcript panel while this runs. It is worth it for two reasons: you get the whole script’s structure at a glance, which is the fastest way to remember what you actually recorded, and you can watch each removal disappear from the text as the edit is applied, so you can tell immediately whether the agent understood the instruction.

The text-based editing workflow maps spoken words to the corresponding timeline ranges, and you can also run the Speech Cleanup Skill for filler and pause removal in ChatCut. Read for missing context, then listen across every cut. A clean transcript can still hide a clipped word or awkward visual jump.

4. Add and style captions

Once the spoken edit is stable, add captions. Specify full sentences or word-level emphasis, plus the intended style.

Built-in caption presets, or the Add captions Skill, are the fastest route. If none of them is what you want, upload a screenshot of a caption style you like and ask ChatCut to reproduce it, or set the type, color, and position by hand in the editor. The AI captions guide covers generation and styling, and the W3C WAI captions guide outlines timing and speaker-identification essentials. Always check proper nouns, product terms, numbers, and line breaks.

5. Finish the A-Roll with sound, transitions, and zooms

Sound effects, transitions, and punch-in zooms are the cheapest way to connect ideas and mark emphasis, and they belong to the spoken edit rather than to the B-Roll pass.

You are not limited to what the agent chooses. ChatCut ships a sound-effect library and a transition library, and you can drag any item straight onto the timeline yourself.

Zooms work the same way: ask for them in the conversation, or add one on the timeline and open its editor to set the framing, the timing, and the easing. If you have ever set keyframes in a conventional editor, this will feel like a much shorter version of the same job.

That completes the A-Roll. The spine of the video now says what it needs to say. Everything from here adds to it.

6. Add Motion Graphics

For a talking-head video, Motion Graphics are the cleanest form of B-Roll: they carry meaning instead of merely filling the frame. Four moments are usually worth one:

• Laying out the structure of the whole piece near the top.
• Marking the start of each point in a numbered or sectioned argument.
• Making an abstract idea concrete when words alone are doing too much work.
• Emphasizing a single line you want the viewer to keep.

Ask for them by purpose and let the agent place them.

ChatCut offers a set of tuned visual directions to choose from, so a whole video shares one look instead of collecting unrelated inserts.

Once a direction is applied, the graphic sits over the footage as its own layer rather than being burned into the picture.

If the result is close, revise its text, colors, spacing, placement, or exposed design properties instead of regenerating the whole video. You can prompt the change, or open the Motion Graphics editor and adjust the palette and type sizes by hand.

You can also start from a built-in template, place it on the timeline, and save a graphic you like for the next project. If you have settled on a full visual system, including a logo and brand fonts, save it as a Design Style and reuse it. The Motion Graphics feature guide shows the broader workflow.

7. Generate the images and video you cannot shoot

Some points need a shot you do not have and cannot license. ChatCut can generate a supporting image or video without leaving the task. Specify its purpose, reference, duration, and placement.

For a still image or thumbnail, define the composition and constraints as carefully as the subject.

Two different meters can apply here. When Codex uses its own image generation, that runs on your ChatGPT/Codex allowance and does not create a ChatCut charge. Generation performed by ChatCut is billed in credits by model, resolution, and length, and AI video generation requires a paid plan. The editor shows an estimate before the job starts, and the credits policy lists the current rates. The AI video generator covers the separate job of creating new footage: editing changes source media you already have, while generation creates a new asset to use within the edit.

8. Add music, voiceover, and sound effects

For music, explain the role it should play rather than asking for a generic track. Mention the pace, emotional direction, and whether speech must stay dominant.

Generated voiceover covers two useful cases: a line no one recorded, such as the other half of a conversation, and creators who would rather not appear or speak on camera at all. Generated speech is transcribed like any other clip, so you can keep refining it with the same text-based editing you used on the A-Roll.

Review pronunciation, timing, and levels, and cut anything that does not earn its place.

9. Review the timeline and export

Watch the complete piece, then inspect its opening, major cuts, captions, graphic edges, audio changes, and final frame. Request specific revisions (“restore the pause after the first sentence” is better than “fix the pacing”) or edit the timeline directly.

When the result is ready, ask the agent to export or open Export in ChatCut. The active timeline can be delivered as a finished video, audio, subtitles, or an XML handoff in FCP7 XMEML for Adobe Premiere Pro and DaVinci Resolve (plus editable CapCut or Jianying drafts on Desktop when those apps are installed locally). Inside the XML export, Include motion graphics renders each graphic as a transparent ProRes 4444 file beside the timeline; that option requires Pro. Check imported timelines either way, because not every visual treatment travels through XML.

Do not treat a completed render job as the final quality check. Play the actual exported file and verify picture, audio, captions, resolution, orientation, and duration. The export guide lists what each delivery type includes.

Copyable prompts for common editing tasks

For a complete first pass, combine the outcome and constraints in one brief, then request targeted revisions. For a repeated format, turn the brief into a reusable video-editing master prompt.

Talking-head cleanup

Long-form video to Shorts, TikTok, and Reels

For a guided workflow that finds complete ideas and formats each vertical clip on its own timeline, open the Long Video to Shorts Skill.

Product demo or explainer

For revisions, point to one timeline, time range, caption, or visual and protect everything else: “At 00:31, replace the callout with a two-column comparison. Keep the cut, audio, captions, and all other Motion Graphics unchanged.”

Beyond talking head: what else people build this way

The A-Roll then B-Roll method holds outside the talking-head format. Only the source and the finishing layers change.

• Long-form to short clips. Ask for complete ideas rather than fixed durations, then check that each clip works vertically. See turning long videos into shorts and social media content.
• Short film and narrative work. The A-Roll is picture and performance instead of speech, so selection matters more than filler removal. See AI filmmaking.
• UGC-style product video. A creator take is the A-Roll; product inserts, captions, and on-screen claims are the B-Roll. See the UGC video workflow.
• Advertising. Short, repeatable, and heavily dependent on a consistent visual system, which is where saved Design Styles pay off. See product ads.
• Explainers. Narration carries the logic while Motion Graphics carry the structure. See making an explainer without After Effects.

Talking head remains the best format to learn on, because the transcript gives the edit a structure you can read. The talking-head editing workflow goes deeper on that case.

What ChatGPT still needs you to review

Before publishing, check:

• Meaning and context. Make sure a shorter quote still represents what the speaker intended.
• Pacing. A transcript may read smoothly while the audio feels rushed or a visual cut feels abrupt.
• Caption accuracy. Confirm names, numbers, terminology, punctuation, and readable line breaks.
• Visual emphasis. Remove B-roll, zooms, or Motion Graphics that compete with the point.
• Rights and disclosure. Confirm permission for source footage, logos, music, likenesses, and generated assets, and follow platform rules such as YouTube’s altered or synthetic content disclosure policy when publishing realistic generated media.
• The exported file. Play it outside the editor and check picture, audio, captions, start, end, format, and orientation.

Why this workflow exists

A complete video workflow can span several specialized capabilities. A chatbot can reason about a transcript. An image or video model can produce a shot. Other tools handle music, sound effects, and narration. The practical problem is getting every useful result into one project where it can be arranged, retimed, revised, and exported.

Two problems follow. The more material you create, the heavier the handoffs get, and you end up back in a conventional editor to arrange it all. And a generated asset tends to be a one-shot result: your only connection to it is the prompt that made it, which is a thin and fragile thing to hold. Getting to a passable output is fast; improving that output is where the time goes.

Coding agents changed the second half of that equation in 2026 by taking instructions, running tools, and reporting back. Connecting more generators to an agent, though, does not by itself produce an edit. The outputs still have to land somewhere they can be arranged, retimed, and replaced.

That is the gap ChatCut fills: a real editor, with a timeline, media pool, player, transcript, and property panels, connected to the agent you already talk to.

A list of suggested timestamps is still a plan. A real edit changes the project and leaves you with a timeline or a finished file.

Troubleshooting ChatGPT video editing

The plugin or Desktop tools do not appear after installation

First verify whether chatcut_desktop (for ChatCut Desktop) or chatcut (for the hosted Web plugin) tools are already available in the session. Because agents load MCP tools when a session begins, start a fresh desktop task after first-time installation or after launching ChatCut Desktop. If tools are still missing, confirm that you signed in inside ChatCut Desktop (the most common cause of missing Desktop tools) and check Agents on this computer in the Desktop sidebar or run codex mcp get chatcut_desktop in a terminal. For the hosted Web plugin, wake the plugin with @chatcut in a new task or copy the current prompt from ChatCut Web avatar → Agent Plugin → ChatGPT/Codex.

The editor does not open

With the default ChatCut Desktop setup, the editor runs as a native macOS or Windows app rather than inside the agent’s Browser panel, and the agent can even edit a selected local .chatcutproject package while the Desktop window is closed (Edit a Desktop Project from Your Agent). Open ChatCut Desktop yourself when you want to inspect the timeline visually. If you are using the hosted Web plugin instead, ask the agent to open the project in its Browser panel, complete browser authorization if prompted, and never paste account passwords, cookies, or access tokens into the conversation.

The desktop app crashes when you add a video

In native ChatCut Desktop, dragging files or folders from Finder or File Explorer into My Assets works normally. If you are using the hosted Web plugin’s embedded Browser panel inside the ChatGPT desktop app, add media through My Assets → Upload or the conversation column instead; during our earlier recorded session, dragging a video directly onto the embedded panel could crash the host app (current host defect status is not verified).

Media upload is unavailable or fails

Open the editor and use My Assets → Upload. Wait until the asset finishes processing before asking ChatGPT to use it. If it fails, confirm that the file plays locally and try a common interchange format such as H.264 MP4 for video, PNG or JPEG for images, or WAV or MP3 for audio. A task attachment is context until you explicitly ask ChatGPT to import and use it.

The first edit is too broad or generic

Narrow the next request. Name the source or timeline, select a time range, state the target length and audience, list what must remain, and describe one change. “Tighten the first 45 seconds but preserve the customer quote and CTA” gives the editor a testable boundary; “make it more engaging” does not.

Export takes time or fails

How an export renders depends on which ChatCut surface you use (Exporting, Web vs Desktop Export). In ChatCut Desktop, video renders locally on your computer (up to 4K or HDR on supported hardware) and saves directly to your chosen folder. Through the hosted Web Agent Plugin, video and audio exports use browser rendering by default in a signed-in supported browser for an explicitly identified project and Timeline ID; cloud rendering (up to 1080p) requires your explicit request or agreement and does not run automatically if a browser export fails. Manual Web exports automatically choose browser or cloud rendering. Keep the editor or browser open during local or browser renders, check progress under the header Export button, let Preparing media… finish uploading for cloud jobs, or use Retry in the export queue.

Method and disclosure

ChatCut publishes this first-party tutorial. We originally reviewed the workflow and linked ChatCut documentation on October 13, 2026, and on October 15, 2026 we added chapter links to our September 17, 2026 filmed editing session so you can inspect the source brief, revisions, and export controls alongside the written guide. The accompanying recorded ChatGPT editing test publishes its source video, prompt, visible result, timing, and limitations.

On November 7, 2026, we audited and updated this guide against current ChatCut and OpenAI documentation to cover the Codex workflow, the default chatcut.io/desktop local MCP setup, the separate chatcut.io/web-plugin/chatgpt hosted Web plugin route, local vs. browser/cloud export behavior, and Free vs. Pro plan boundaries. We did not run a new timed runtime benchmark for this update. Plugin behavior, generation models, plans, and usage limits can change, so check the linked product pages and in-app estimate before starting a large project.

The useful division of labor is simple: let ChatGPT or Codex translate your brief into concrete editing and generation work, then use ChatCut’s AI video editor to judge and refine the result on the timeline.