How a clip is made
Five steps, all of them on one machine, with a check before anything leaves it.
The concept
A language model drafts what happens in the clip: the object, the material, what acts on it, and how the action is spread over the seconds available. It also writes the sound description, because picture and audio are generated together.
Generation
An open video model generates the clip together with its soundtrack, locally on one graphics card. Longer clips are produced as consecutive blocks, where the last frame of one becomes the starting frame of the next, so the motion continues instead of restarting.
Joining and scaling
The blocks are joined with a short crossfade and scaled to the vertical format the platforms expect.
Technical check
Before anything is published, an automated check confirms duration, file size, resolution, aspect ratio and that a soundtrack is present across the whole clip — not just at the start. Anything that fails is not published.
Publishing
The finished file goes to the channel's own accounts through each platform's official interface, with the AI disclosure set where the platform provides one. Nothing is posted to any account that does not belong to the channel.
What runs where
On the local machine
Video and audio generation, joining, scaling, the technical check, and all credentials.
External services
A language model interface for drafting concepts, private object storage to hand a finished file to a platform, and the platforms themselves.
Not involved
No analytics, no tracking, no data from anyone else. There are no other users of this tool.
Honest about the limits
Generated video is not reliable. Clips come out with wrong physics, with the object turning into the wrong material, or with sound that does not match what is on screen. Those are discarded. The check catches the technical failures; the rest is a matter of looking at the result and deciding whether it is worth publishing.