AI Video Editor

The AI video editor edits footage you already have. You upload a video, type what should change, and Claude makes the edit. There is no timeline to learn and no clips to drag: an instruction such as "cut the pauses and add captions" is the whole interface. It is the sister tool of Code to Video, which writes a new video from a description; the editor starts from your own recording instead.

Under the hood, Claude writes the edit as code. Your video plays as clips with a start time, a length, a point in the source and a speed. Captions, titles, a logo and color grades sit on layers above it, and one animation timeline moves them. The renderer opens that page in a headless browser and records it frame by frame, so what you preview is what you get.

Claude does not guess. When you upload a video, the speech is transcribed with a time on every word, so cuts land between words and captions match the voice. Your browser also draws a contact sheet of twelve frames across the video, each marked with its time, so Claude sees the framing, the people and any text already burned into the footage.

That is why an instruction like "end after she says welcome" or "keep the speaker in the frame" works. Claude reads the words, looks at the frames and places the cuts and the crop to match.

Say the result you want, not the steps. "A 30 second vertical short with the best answer, bold captions and a title at the top" works better than a list of cuts. Name the length, the format and the platform if they matter.

Refer to the speech when you can. "Start at the question about weather" or "cut the part where he repeats himself" is precise, because Claude has the transcript. You can write in English, Chinese or any other language, and ask for captions in another language than the speech.

The first result is a draft. Ask for the next change in plain words, such as "make the captions bigger", "only the first half in black and white" or "add the title for the first three seconds". Claude keeps everything you did not mention and changes only what you asked for.

Every round becomes a version. Go back to an earlier one, branch from it or start over. If you know HTML, CSS or GSAP, you can also open the code and change a value by hand before you render.

After each edit, the real renderer takes four stills of the result. Claude compares them with your instructions and fixes serious problems, such as a caption over a face, text cut off at the edge or a black gap between clips, before you see the result. Writing the edit and the previews are free: three edits a day before your first purchase and thirty a day after it.

Rendering the MP4 uses purchased credits: 15 credits up to 30 seconds, 18 up to 40 seconds, 22 up to 50 seconds, 26 for one minute and 50 for the two-minute maximum. A failed render is refunded.

Upload MP4, MOV or WebM video up to 5 GB and 10 minutes, MP3, M4A or WAV audio, and PNG, JPEG or WebP images, up to five files per edit. The output is a 1080p MP4 in 9:16, 16:9 or 1:1, up to two minutes long. Your uploads are stored in a folder of your account under random names and are used only for your edits.