Blog · August 2, 2026
Can ChatGPT Edit Videos? What It Can and Can't Do in 2026
TL;DR: No. ChatGPT cannot edit videos on its own — it has no timeline, no timecode, no access to your media, and video isn't an accepted upload type. It can't generate video either: OpenAI discontinued Sora on April 26, 2026, and the API shuts down on September 24, 2026. What it can do, since MCP connectors arrived, is drive an editor that does have those things — ChatGPT writes the edit, a separate app performs it.
What happens when you hand ChatGPT a video file
Nothing, mostly. Video isn't on the list of file types ChatGPT accepts. You can upload images, PDFs, Word documents, presentations, spreadsheets and plain text — a .mp4 or .mov gets rejected before the model ever sees it. Your footage doesn't make the cut. (It won't be the last time that joke applies to this post.)
I spent a while assuming I was holding it wrong. Reader, I was not holding it wrong.
Even if the file cleared the door, the tooling isn't there. Editing video means four things ChatGPT has none of:
- A timeline. No sequence, no tracks, no in and out points.
- Access to your media. Your footage lives on a drive or a phone. ChatGPT can't reach it.
- Timecode. Without frame-accurate positions, "cut at the pause" isn't an instruction anything can execute.
- A render pipeline. Nothing on the other end to encode and write a new file.
The common workaround is pasting a YouTube link. Sometimes ChatGPT retrieves a public video's transcript; often it returns metadata and little else. Either way it isn't watching anything. It will never catch the shaky handheld shot or the audio drop at 4:12 — the problems that actually drive an edit.
What changed in 2026: ChatGPT can't make videos either
This is where most articles on this keyword are now out of date. Until spring 2026 the honest answer was "it can't edit video, but it can generate it" — Sora lived inside the ChatGPT surface, and you could describe a shot and get one back.
That's over. OpenAI wound Sora down in two stages:
| Date | What happened |
|---|---|
| March 24, 2026 | OpenAI announces the wind-down and notifies API developers |
| April 26, 2026 | Sora web and app experiences discontinued |
| September 24, 2026 | sora-2, sora-2-pro and the Videos API shut down |
OpenAI's own deprecation notice lists no recommended replacement. Reporting at the time put the reason down to economics — video generation costs far more per output than text, and the compute was redirected toward coding and enterprise products.
Most of the internet has not received this memo. A good half of the pages ranking for "ChatGPT video generator" are cheerfully selling you a product that has been dead since April, which makes them less articles than séances. Check the date on anything you read about this, including this.
What ChatGPT is genuinely good at for video
All of it is language work, which is exactly what the model is for:
- Paper edits. Give it a timestamped transcript and it will pick the segments worth keeping and order them into something that holds together. This is the big one.
- Scripts and hooks. Ten opening lines for the same video, then a pass to tighten the one you pick.
- Titles, descriptions and chapter markers. Fast, and it will generate twenty variants to test.
- B-roll and shot lists. Read it the script, get back a list of cutaways to film or find.
Note what these share: text in, text out. The moment the task requires seeing or moving pixels, you're back to an editor.
How to actually edit video with ChatGPT, step by step
Here's what changed. ChatGPT reaches outside tools through MCP connectors, and several editors now publish one. You're not making ChatGPT into an editor. You're letting it drive one.
The division of labour is strict, and it's worth knowing before you start:
| ChatGPT does | The connected editor does |
|---|---|
| Reads the timecoded transcript | Holds the media files |
| Finds the story and orders the beats | Maintains timecode |
| Writes the cutting decisions in plain language | Builds an FCPXML, XML, OTIO or EDL |
| Argues with you about them | Relinks camera files and renders |
- Connect an editor. In ChatGPT's connector settings, add the editor's MCP endpoint. Eddie AI publishes one at
mcp.heyeddie.ai/api/mcp, no auth required, and it's the route I'd start with. - Get the footage in. Upload to the editor, not to ChatGPT. The editor transcribes and indexes it, and ChatGPT reads that index rather than your files.
- Ask for the edit in plain language. "Cut this 42-minute interview to nine minutes about how she started the business. Use her exact words. Tell me why each clip earns its place." Specificity beats politeness.
- Watch it, then argue with it. The first pass will keep a line that was delivered flat and drop one that landed. It's working from words. It can't hear tone, and it has no idea take three was soft. (Think of it as a very confident intern who has read the transcript but not watched the tape. Enthusiastic. Wrong in specific ways.)
- Take it into your NLE. Export the sequence to Premiere Pro, Final Cut or DaVinci Resolve, where your original camera files relink and the cuts arrive as clips you can still trim.
That last step is the ceiling, and it hasn't moved. ChatGPT produces a plan. Something else performs it. MCP didn't change that — it just shortened the walk between the two, like a longer arm on a claw machine.
Where this workflow falls apart
It needs someone to be talking. Every step above runs on a transcript, so footage with no dialogue gives it nothing to work with. Point it at a silent montage and it will sit there like a sommelier handed a glass of water.
Say you're editing on your phone — twenty clips from a weekend, no speech, half of them shaky. There's no transcript to index. Nothing in the workflow addresses the actual work: finding the usable seconds inside each clip, cutting the dead air, matching pace to music, exporting vertical.
Then there's the loop itself. On a 40-minute interview, the connect-index-prompt-review-export cycle earns its keep. On a 45-second clip it's several minutes of admin around an edit you could have done by hand in two — the video equivalent of driving to the gym to use the stairs.
What the ChatGPT editing stack actually costs
Nobody ranking for this query mentions that this is a two-subscription workflow. You need a ChatGPT plan that supports connectors, and you need the editor on the other end. Neither is the free tier.
ChatGPT's plans — Free, Go, Plus and Pro — are on OpenAI's pricing page. The figures didn't render when I checked on August 5, 2026, so read them yourself rather than trust a number I couldn't verify.
The editor side is what surprises people. Eddie AI, verified on its own pricing page on August 5, 2026:
| Plan | Price | What you get |
|---|---|---|
| Pay As You Go | $0, credits at $15 each | No subscription, export when you need to |
| Pro | $167/mo billed yearly | 120 exports a year |
| Pro+ | $333/mo billed yearly | 300 exports a year |
| Ultra | $1,250/mo billed yearly | 1,200 exports a year, 5 seats |
Read that middle column again, because the "/mo" is doing a lot of heavy lifting. $167/mo billed yearly is $2,004, once, upfront. The page opens on the yearly toggle; monthly runs about 16% higher. This is the category's favourite trick and it is worth naming every single time: when a tool advertises a monthly number that requires twelve months of commitment, that is not a monthly price, that is a lay-by with better branding. Quote both figures or quote neither.
Credit where it's due on the other axis, though — the plans bill per export, not per minute of source footage. That's the more honest of the two models. A 90-minute interview and a 90-second clip cost the same to finish, so the footage that most needs the help isn't the footage that punishes you for uploading it.
This is professional tooling priced for professionals, and if you cut multicam interviews for a living it will pay for itself. If you shoot on a phone, you are in the wrong aisle entirely, holding something with a chainsaw where the scissors should be.
What happens to your footage once you connect a tool
The other thing nobody on page one mentions. Connecting ChatGPT to an editor means your footage — or at minimum its full transcript — is being indexed somewhere, and "somewhere" is worth pinning down before you upload a client's interview.
Ask any tool what happens to your footage after processing. Most pricing pages don't say, and the silence is the answer. A privacy policy that takes four hundred words to avoid the question has, in fact, answered it.
Eddie AI's does say, to its credit: it runs as a desktop app, the footage stays on the local machine, and it doesn't train on your media. That's a straight answer, which in this category counts as a feature.
Ours, for the same reason: with 1 Tap Cut, short clips are processed on-device and never leave the phone. Longer projects go to the cloud encrypted and are deleted within 24 hours. We never train on user video. "Nothing ever leaves your phone" would be false, so I'm not going to write it.
What to use instead, by footage type
It splits by footage, not by brand:
- CapCut — the default for short-form, free, deep template library. Also the reason a scroll through any feed shows the same six transitions.
- Premiere Pro / DaVinci Resolve — multicam, long-form, colour work. Resolve's free tier is genuinely capable and costs nothing.
- Descript — edit the video by editing its transcript. The closest thing to editing with words.
- Eddie AI — the ChatGPT-connected route above, aimed at working editors with dialogue-heavy footage.
- OpusClip — pulls short clips out of long uploads, priced per minute of source video.
- 1 Tap Cut — cuts phone footage into a finished vertical edit on the device, without a timeline.
Don't use 1 Tap Cut for a 40-minute multicam interview. There's no timeline and no multicam. Open Resolve, transcribe it, and let ChatGPT write you the paper edit — that's the job it's good at. We've put verified pricing for several of these side by side in our OpusClip vs Submagic vs Descript comparison, because the billing model matters more than the feature list once you're a few months in.
FAQ
Can we edit a video in ChatGPT?
Not inside ChatGPT itself. It has no timeline, no timecode and no access to your media, and video files are not an accepted upload type. Connected to an editor through an MCP connector, it can write the edit while the editor performs it.
How do you edit video in ChatGPT?
Add a video editor's MCP connector in ChatGPT's settings, upload your footage to that editor, then describe the cut you want in plain language. ChatGPT reads the timecoded transcript and returns cutting decisions; the editor builds the timeline and exports it to Premiere Pro, Final Cut or DaVinci Resolve.
Why can't ChatGPT make videos?
Video generation in ChatGPT came from Sora, which OpenAI discontinued. The web and app experiences ended on April 26, 2026 and the Sora API shuts down on September 24, 2026. OpenAI's deprecation notice lists no replacement model, and reporting at the time attributed the decision to the cost of video generation compared with text.
Can you upload a video file to ChatGPT?
No. ChatGPT accepts images, PDFs, documents, spreadsheets and text files, but not video. The workaround is to transcribe the video elsewhere and paste the timestamped transcript, or to connect an editor that holds the media for you.
Can AI edit videos automatically?
Partly. AI reliably automates specific passes — removing silence and filler words, generating captions, finding clip-worthy moments, reframing to vertical. Choosing which take is better, and why, is still a human call, because the model can't hear that a line was delivered flat.
What is the best prompt for video editing in ChatGPT?
State the source length, the target length, the subject, and the output format. For example: "Here's a timestamped transcript of a 42-minute interview. Cut it to nine minutes about how she started the business. Return a table with start timecode, end timecode, the quote, and one line on why it earns its place. Don't paraphrase."
Can ChatGPT create videos for free?
No, on any tier. Video generation was removed with Sora in April 2026, so no ChatGPT plan generates video. Paid plans are needed for connector access, and the connected editor is a separate subscription.
Which AI video editor is best for beginners?
CapCut, if you want templates and a free tier. For phone footage cut into vertical clips without a timeline, 1 Tap Cut is built for that specific case and is currently in closed beta. Both apply a watermark on their free tiers, which is the real limit on any free plan in this category.
Is video editing dying due to AI?
No. The first pass is being automated — logging, rough assembly, captions, silence removal. Deciding what the video is about, and which take carries it, hasn't moved. The tools that connect ChatGPT to a timeline are explicitly built as assistants to editors, not replacements for them.
The short version
ChatGPT is a writing tool that happens to be useful either side of an edit. It plans, it scripts, it names things. Give it a connector and it will tell a real editor what to do, in the manner of a director who has never touched a camera.
It still does not cut. Which, for a thing half the internet keeps calling an editor, is quite the oversight.
If your footage comes off a phone and you want it cut, captioned and formatted for TikTok, Reels or Shorts without opening a timeline, that's what 1 Tap Cut does — on-device, exporting vertical, square and widescreen from one edit. It's in closed beta; join the waitlist and invites go out weekly. See pricing for plans.
And if you came here hoping to drag an MP4 into a chat window and get a finished vlog back, I'm sorry. Genuinely. We all wanted that one.
Sources: Sora deprecation dates and model names from OpenAI's API deprecations page and the-decoder's report on the two-stage wind-down. MCP connector endpoint and Eddie AI pricing verified on heyeddie.ai/pricing on August 5, 2026. Post updated August 5, 2026.