Start and end frames

A start frame is the picture the clip begins on. An end frame is where it is heading. Give a model both and you have described the movement without writing a word about it, which is useful because movement is the hardest thing to put into a sentence.

This is the ingress to reach for when the two ends matter more than what happens in between: a room before and after a restyling, a product turning to face you, a face going from neutral to laughing.

Some models read a start frame and decide the ending themselves. Some take both. A few take only the pair and cannot be run from a prompt alone. The studio shows the ports a model actually has, and moves you to a variant that reads what you attached.

If you attach an end frame to a model that has no port for it, you get told, rather than having it quietly dropped. A reference that goes missing without a word is the expensive kind of mistake: the run is paid for and the result is not what you asked for.

The two frames should come from the same world. Two stills with different light, a different lens or a different room give the model a contradiction to resolve, and it resolves it by inventing a cut.

Both can be made in the studio, on an image model, before the video runs. That is the cheap half of the job: an image costs a fraction of a clip, so it is worth getting the two ends right before you pay for the movement between them.

On the canvas this is one pipeline. An image generator feeds the start frame, a second feeds the end frame, and a video generator reads both. Build it once and the next product goes through the same shape.