Why every clip starts on the last frame

A video model makes a few seconds at a time. If you want a minute, you need a way to join many of those seconds so that the join does not show. AI Vids has one simple rule for it.

The rule#

Each clip starts on the last frame of the clip before it. The first clip starts on a picture of your own, or from words alone. When it is made, its last frame becomes the first frame of the next clip, and so on until the video is as long as you asked. Then the clips are stitched into one file.

What it gives you#

Continuity of the picture: the next clip begins exactly where the last one ended, so there is no jump in colour, framing or position at the join. That is what lets a video run for minutes without a visible seam.

What it does not give you#

A story. The chain keeps the picture continuous, but what each clip shows is decided by its prompt and the model. A chain of clips with prompts that do not belong together will be continuous and confusing. Write each clip's prompt as the next beat of the same scene, and use a cut when you want to start fresh.

Cuts and end pictures#

A cut starts a clip fresh instead of on the clip before, for when you want a new scene. You can end a clip on a picture where the model allows it, so a clip lands exactly where you want, and the next one starts there.

With the Previews#

Today's default models are Previews that paint clips in your browser. They honour your start and end pictures exactly, because frame 0 is your picture and the last frame is your end picture. They are a good place to learn the chain, and the clips are simulated.

With real models#

A real model makes less predictable clips, and the last frame may not be where you hoped. The timeline lets you remake one clip with a new seed without touching the others.

See the clip chain.