How to Remake a Video with AI

What you need before you start

One video file, MP4 or MOV, within the length and size limits shown on the upload box. It can be something you shot yourself, something you have the rights to, or a clip you are using as a reference for blocking and pacing. Nothing else — no script, no character descriptions, no shot list. Producing those is the whole point of the read-through.

A fast-cut clip has more shots to read than a slow one of the same length, so it takes longer — but it does not cost more. The read-through is priced by the length of the clip alone, rounded up to the next minute; the number of cuts never enters the price.

Step 1 — Upload the clip

Open the home page, switch to the Remake tab, and pick your file. The upload starts the moment you choose it, so the price appears before you commit to anything. If you change your mind, remove the clip and the storage is handed straight back.

You can also type a sentence about the clip. It is a hint, not an instruction — something like “the woman in grey is the lead, the two men are brothers”. The model cannot tell that from pictures alone, and where the sentence disagrees with what is on screen, what is on screen wins.

Step 2 — Watch the read-through

This is the part that takes a few minutes. It runs in four passes, and the progress screen names each one as it happens: find the cuts, identify the cast, read every shot, then check and save.

What comes out the other end is a complete project: every person who appears becomes a character with a written appearance and a voice description, every place becomes a location, every prop that matters becomes an item, and every shot becomes a numbered beat inside a segment — with its framing, its camera move, who is in frame, and the line that is spoken.

You can see exactly what that looks like in the sample episode: the left column is the cast the read-through produced, the middle column is the storyboard, and each coloured chip is a real link back to a character, a location or a prop.

Step 3 — Choose how it should look

The read-through tells you what happens. You decide what it looks like. On the first step of the wizard you pick the render style, the art direction, the wardrobe, and a target region if you want the cast moved to a different part of the world. Ethnicity, faces, architecture and street signs follow the region you pick; names do not — they are already written into the storyboard the read-through produced, so renaming a character is a manual edit in the next step.

This is the point where a remake stops being a copy. The beats are the source clip's; the people, the clothes, the architecture and the language are yours.

Step 4 — Generate and composite

From here it is the normal pipeline: generate the character sheets, then shoot segment by segment, then composite the episode. Each button shows what it costs before you press it. Note that the free allowance a new account gets on its first project does not apply to remakes — a remake is charged at list price from the first setup image.

Segments are short on purpose — four to fifteen seconds each. That is one generation call, and it is also the window inside which a video model can hold a face steady.

What a remake will not do

It does not copy the original footage. The source clip is read and then set aside; it is never handed to the video model as a reference frame. That is deliberate: a model given a reference frame copies the faces in it, and copying a real person's face is the one thing we will not build.

It does not reuse the original audio. The lines are heard, then rewritten in your project's language and spoken by your own cast's voices.

It is not a licence. Reading a clip does not give you the rights to its story. Use footage you own, footage you have cleared, or footage you are only using to study timing.

FAQ

Updated October 2, 2026