Create

Video, shot by shot

Go from image to film, shot by shot, starting from built images. Each shot is judged on its own and redone on its own; the edit is only made from what is validated.

Video, shot by shot

A film is not an image that moves. It is a sequence of shots, and each one has its frame, its duration, its camera move, its match with the previous one. If you ask for everything at once, you get everything at once, and you can take nothing back. At Axiris, the film is built like the image: shot by shot, and what is validated no longer moves.

The studio timeline

In the Axiris studio, a film is composed on a timeline. Each shot carries its start image, its written intention, its camera move on its own line, and up to nine reference images: the character, a setting, an object. The ambient sound is directed shot by shot.

A shot is launched on its own. You look at it, you keep it or you redo it with another seed, another intention, another reference, without touching the neighbouring shots. The edit assembles what has been validated, and only that. A nine-shot storyboard, rendered from a single character sheet, gave a film in which the character's identity holds from one shot to the next: this is judged by eye, and we say so.

This way runs in the studio on an open-weights model whose licence limits the countries where it can be used: MiniMax H3. The tool: Timeline Director.

The whole sheet, through the cloud

The video models of the major services take a whole storyboard sheet as input and render the film in one go, cuts included. It is fast, it is the model that decides where the cuts fall, and it is usage that you pay for. Axiris is preparing this way in the same interface, with Nova to advise which one to take and how to talk to it: Seedance Pro, Cloud of your choice, with Nova.

The two ways are not opposed. You pull a whole sheet to see quickly; you switch to shot by shot when you need to hold a character, a cut, a match.

What makes the quality of a film

The quality of the images you give it. A video model adapts what it receives all the way through: an accurate start image gives an accurate shot. That is why the image chain is upstream of the video: redo an area on the start image, place an exact element in it, and it enters the shot.

One reference image, one body. A character sheet is given view by view, never as a pasted board, otherwise the views end up in the setting: Character sheet.

What we do not promise

An element pasted flat, which has no business in the scene, does not carry through the video: the model does not keep what it does not believe in. And an identity resemblance measurement does not exist yet: when we say the character holds, it is an eye that says so.

To talk about a video project: Contact request.