Edit video with AI: type the change, check it on a timeline

Edit video with AI: what GPTVideoEdit actually changes

GPTVideoEdit is a browser video editor with an AI assistant beside the timeline. You upload a video you already have, type the edit you want in plain words and the assistant turns it into timeline changes: a cut, a timed title, a mute, a speed change, a new frame shape or a zoom. Every change lands on an editable timeline, so you can adjust it by hand or undo it before you export.

The AI plans the edit and your browser carries it out on your own footage. Nothing is regenerated: a request to remove a pause removes that range from your recording, and a request for a title adds a text layer over it. That is the difference between an AI video edit and AI video generation, and GPTVideoEdit only does the first.

GPTVideoEdit is independent and is not an official ChatGPT, OpenAI or Anthropic product. Quick and Standard use OpenAI language models and Advanced uses an Anthropic model. A ChatGPT subscription does not add credits here, and you do not need a ChatGPT account to use the editor.

Make your first AI video edit in five steps

Pick a short clip for your first try and ask for one change at a time. You can trim, split, arrange and export by hand without an account; signing in is only needed when you want the AI assistant to do the work.

1. Bring in the video

   Drop a file on the upload area or browse for it. The beta accepts files up to 1 GB and projects up to 15 minutes. Your browser has to be able to decode the file, so if a video will not open, an MP4 with H.264 video and AAC audio is the safest format to try.

2. Say what should change, where and when

   Give the action, the part of the video and the time. “Cut 00:21 to 00:25” or “Put the word Recipe at the top for the first four seconds” leaves nothing to guess. Select a clip first when the change is about that clip’s sound, speed or framing.

3. Choose Quick, Standard or Advanced

   Each request shows the most credits it can reserve before it runs. Whatever the edit does not use is returned when it finishes. The model changes how the edit is planned, not what the editor is able to do.

4. Watch the result on the timeline

   Play the part that changed with a few seconds either side. Check the words, the sound and the moment of each cut. Fix small things with the timeline controls, or undo and ask again more precisely.

5. Export and check the file

   Export to MP4 or WebM with a maximum height. The editor never upscales beyond your source. Leave the tab open until encoding ends, then play the downloaded file from start to finish.

How to word an AI video edit so it comes out right

Strong requests name four things: what to do, what to do it to, when it happens and what must stay as it is. “Make it better” gives the assistant nothing to act on; “Lower the selected clip to 80% volume and leave the picture alone” gives it everything.

Times always refer to the timeline as it is now. When you cut a section, everything after it moves earlier, so a timestamp you noted from the original recording may point somewhere else after your first edit. Read the playhead before you write the next request.

Keep different decisions in separate requests. Cut the false start, watch it, then decide on a fade. Add the title, look at it over the footage, then ask for a bigger size. Small steps make it obvious which request caused which change and make undo more useful.

Requests you can copy for cuts, titles, sound, framing and zoom

Swap in times from your own timeline. After each request, watch slightly before and after the change: an edit can be technically correct and still land in the middle of a word or a movement.

1. Open with a title

   “Show Five minute pasta in white with a dark background at the bottom from 00:00 to 00:04.” Make sure the title does not cover a face or something the viewer needs to see.

2. Cut a known mistake

   “Remove 00:31 to 00:36 and close the gap.” The project gets five seconds shorter, so later times move with it.

3. Tighten the pauses

   “Cut the long silences but keep everything that is said.” The editor checks the audio on your device for quiet ranges. Quiet is not always unwanted, so listen to each join.

4. Punch in on a moment

   “Zoom the selected clip to 130% from 00:10 to 00:13 as an instant punch-in.” Punch-ins keep the sound and timing; check that the subject stays in frame at the higher zoom.

5. Make it vertical

   “Change the project to 9:16 and keep the whole frame visible.” A centred crop fills the frame instead but can cut off anyone near the sides; the editor does not track a moving subject.

6. Silence a few seconds

   “Mute 00:15 to 00:18 without cutting anything.” The picture keeps playing and the sound returns after the range.

7. Speed up a slow part

   “Play the selected clip at 1.25 times speed.” In this beta, changing speed also changes the pitch of voices.

8. Fade the ending

   “Fade the picture out over the last second of the selected clip.” Picture and sound fades are set separately at each end of a clip; they are not cross-dissolves between two clips.

The timeline is where you check and finish every AI edit

Whatever the assistant writes in the chat, the timeline shows what really changed: new clip edges, a removed range, a text layer with a start and end, a mute interval or a zoom range. Check those rather than the summary, and use undo to compare with the step before.

You can do any of it yourself too. Split at the playhead, drag a clip edge, move a title or change its timing without spending credits. Every track takes video, audio, images and text; tracks higher up sit in front, and audio from every active track is mixed together. Images start at five seconds and can be stretched. Deleting a clip leaves a gap, while removing a time range closes it across all tracks. With nothing selected, the side panel shows your media library, and you can drag media from it onto any track.

What the assistant knows about your video

The assistant works from the project: clip lengths and order, text layers, the selected clip and your words. It does not watch your recording and it does not get a transcript of the speech, so timestamps from you are the most reliable way to point at a moment.

For pause requests, quiet ranges found by analysing the audio on your device are passed along as evidence. Automatic transcription and captions, object removal, generated footage and generative effects are not part of this beta.

When an AI video edit is not what you expected

Wrong clip changed? Undo, select the clip you meant and ask again. Wrong time range? An earlier cut has probably shifted the timeline, so read the current time and repeat the request with it.

Text missing? Put the playhead between the layer’s start and end, and check that its colour stands out from the footage. Sound gone? Look at the clip volume and any mute ranges, then play outside them to find where audio comes back.

Video will not import? The file extension does not tell you the codec inside; try an H.264/AAC MP4 that plays normally elsewhere. Export stopping? Keep the tab in front, choose a lower maximum height or shorten the project. Memory and the encoders your browser has decide what can finish.

Export MP4 or WebM and keep your source files

MP4 uses H.264 video and AAC audio and WebM uses VP9 and Opus, where your browser supports them. The height you choose is a limit, not a quality boost. Check the final file in the app or site where you plan to post it, especially after changing the shape.

Encoding runs in your browser. Supported browsers write long exports straight to a file you pick; the in-memory download used elsewhere is limited to projects of five minutes. Being able to preview a video does not guarantee your browser can encode every format.

Projects and the original files can be kept in this browser on this device. A project backup holds the edit instructions and file identities, not the footage, and saving to your account stores the same kind of metadata. Keep your originals somewhere safe, because clearing browser data can delete local copies.

Credits, plans and what an AI video edit costs

Quick, Standard and Advanced choose which model plans the edit. All three use the same timeline tools and the same export engine, so a pricier model does not raise the resolution limit or unlock effects the editor does not have. The estimate on each request tells you the most it can cost.

Editing by hand and exporting never use credits. Paid plans include a monthly AI allowance; annual plans are paid up front and still refresh monthly, and included credits do not roll over. Extra credits you buy do carry over and need an active video plan. GPTVideoEdit and GPTVideoEditor share one video account, plan and balance; photo plans from GPTPhotoEditor are separate.

Frequently asked questions

Video editing guides by task

Follow a complete workflow for the change you need, with example prompts, review steps and advice for the exported file.