Can ChatGPT Edit Videos? What It Can and Can't Do (2026)

ChatGPT has no video editor of its own. You can attach a video file, even on the Free plan, and it may look at parts of it, but OpenAI's help center says it "may not analyze the entire video or accurately interpret its audio," and OpenAI documents no way to edit a video and hand it back. It can't change the clothes, the background or anything else inside your footage either, and OpenAI's own video model is gone: the Sora app closed on April 26, 2026 and the Sora API on September 24, 2026. Plugins from other companies, such as Adobe and Descript, can trim, caption and resize a clip inside ChatGPT. Changing what is in the shot takes a tool that renders the clip again. We make one, Riffkit, and test it below; on any ChatGPT plan, including Free and Plus, you can use it in Riffkit's web app.

What ChatGPT does with a video by itself

Checked against OpenAI's help center on October 30, 2026:

Video attachments count toward your upload limits: Free accounts can upload 3 files a day, and any plan up to 80 files every 3 hours. OpenAI's data analysis page lists spreadsheets, PDFs and text files, not video, though one hands-on write-up (Atlas Cloud, October 23, 2026) had the Python tool run FFmpeg to turn a clip into a GIF. ChatGPT's image tools can edit a still frame, but OpenAI documents no way to put an edited frame back into your video.

Plugins that edit video inside ChatGPT

OpenAI's Plugins directory is where the editors are. What the main ones do with a video, from their listings and help pages on October 30, 2026:

CapCut said on October 27, 2026 that you'll be able to take an idea from ChatGPT into a first cut in CapCut "in the coming months"; it isn't live yet.

Pick by the job. To trim, caption or reformat footage, use a cutting plugin; to change what is in the shot, you need a tool that renders the clip again.

Changing what's in the shot: our test

Riffkit's Swap renders your clip again from a written description of the change, and aims to keep the person, the movement, the camera and the timing. We ran this test in Riffkit directly, through the same system the ChatGPT connection uses, not inside a ChatGPT chat.

The clip. One we made with AI, 5 seconds long: a woman in a navy suit sprints across an office plaza toward a low camera, under a glass canopy and glass towers, with the caption "me escaping 47 'where's my order' emails" and music, no speech. It's the clip we broke down in Can Claude analyze videos?

What we asked for. Change only her clothes: the navy suit jacket and shirt become a long camel-colored wool coat, worn open, over a plain white crew-neck knit sweater. Keep everything else exactly as it is: her face, hair, run, the setting, the camera, the timing, the sound and the caption. The coat and sweater plain, with no lettering, and no bag or scarf.

Engine and cost. Seedance 2.0, the video model we picked, at 720p, on October 30, 2026. It cost 720 credits: the source file is 5.04 seconds (its sound runs a little past the picture), billed as 6, at 120 credits a second. The render took about 4 minutes, because Riffkit had already analyzed this clip for a day-to-night Swap on October 26; that time, analysis and render together took about 10 minutes.

The coat is there in every frame she's in. Camel wool, open, over a pale knit sweater, from her first step into the frame to the last; we looked at every fourth frame of the 120. Between 3.2 and 4.2 seconds, where the clip has quick jump cuts and a fist pump, we stepped through both videos an eighth of a second at a time: her face, her head position and the fist line up with the original at every step.

What stayed. Her face and panicked expressions, the claw-clip bun and the loose strands of hair, the thin gold chain (now over the sweater), the glass canopy, the towers, the grey sky, the low camera, the framing, and the caption's words and position. The soundtrack is the original: set against the source, it matches with no offset.

What drifted.

• The sweater came back cream and ribbed, not plain white.
• Her trousers went from navy to black, although the brief asked to keep everything else.
• The caption was drawn again by the model: same words in the same place, with slightly bolder, tighter letters.
• The file is 0.04 seconds shorter: every frame is there, and only the bit of sound that ran past the picture was cut.

What to take from it. In this one test, on a 5-second AI-made clip of one person, the outfit change held across the whole clip. Name the pieces you're keeping as well as the one you're changing (trousers, shoes, a bag), read any on-screen text afterwards, and try a short clip of your own before a long one. For several outfits in one clip and how to write the brief, see how to change an outfit in a video with AI; for a new setting, how to change a video background.

How to do it, with or without ChatGPT

Any account, including ChatGPT Free and Plus. Open the Riffkit web app on a computer or in your phone's browser, upload the clip, choose Swap and write the change. You see the price in credits before anything renders.

From inside ChatGPT. A custom MCP server is a tool you add to ChatGPT by pasting its web address, instead of picking it from the Plugins directory. As of October 30, 2026, Riffkit's ChatGPT plugin is waiting on OpenAI's review, so for now Riffkit is added that way, at https://mcp.riffkit.ai/mcp. OpenAI's help page on custom MCP servers, updated around October 28, 2026, lists:

• Business, Enterprise and Edu workspaces on ChatGPT web: full support, including actions that start work, such as a render. An admin or owner adds the server (on Enterprise and Edu, also members an admin allows).
• Pro: read and fetch permissions only, which covers looking things up but not starting a render.
• Plus, Go and Free: not mentioned. In Plugins, select the plus button; if Add custom MCP server isn't there, use the web app.

If you see Add custom MCP server:

1. In Plugins, select the plus button, then Add custom MCP server.
2. Name it Riffkit and enter https://mcp.riffkit.ai/mcp.
3. Keep OAuth as the sign-in method, tick I understand and want to continue, and select Create as a plugin. ChatGPT sends you to riffkit.ai to sign in; approve the connection.
4. In a chat, type @ and pick Riffkit. Ask for an upload link and add your clip on Riffkit's upload page or in the upload box that appears in the chat, not as a chat attachment. The link works for 90 minutes.
5. Describe the change. Riffkit's tools are built to show the credits first and start the render only after you say yes; you get a link to the finished video, and it also stays in your Riffkit library.

Something to ask once it's connected:

Make me an upload link for my clip, then change only my jacket to a long camel coat. Keep my face, my movement, the background, the caption and the sound.

Riffkit also works in Claude; how to make videos in ChatGPT and Claude has that setup.

What it costs, and what's free

• Billed on your source clip's length, rounded up to whole seconds, 4 seconds at least. A Swap on Seedance 2.0 is 60 credits a second at 480p and 120 at 720p; on Seedance 2.0 Fast, 50 and 100; on Seedance 2.5, 90 and 180. Our 6 billed seconds at 720p came to 720 credits, about $3.60 on Lite.
• The price comes first. In the web app, ChatGPT or Claude, you see the credits before anything renders, and a failed render isn't charged.
• Free to start. No card: new accounts get signup credits. Free videos render at 480p, on Seedance 2.0 Fast or Seedance 2.0, with a small watermark; our 5-second clip quotes 300 credits on 2.0 Fast or 360 on 2.0 at 480p. Clips under 8 seconds come back whole; on a longer video, a free account makes the first section with its signup credits and a plan finishes the rest.

Plans start at $15 a month for 3,000 credits.

Limits to know

• ChatGPT doesn't render anything. Riffkit sends the job to the video model (Seedance, in our test) and sends back a link.
• Short clips. Currently up to 45 seconds and 100 MB.
• One continuous shot carries a change best. Our clip's jump cuts, all from one camera, kept the new outfit. With many camera angles, check each shot: in our day-to-night test, Seedance 2.5 carried the change through more of a 12-shot clip than 2.0.
• Use footage you own or have the rights to. A Swap keeps the person and the sound from your clip; where the line sits covers the rest.