A whole story in one video, scene by scene

A video model makes a few seconds at a time. Ask for three clips with the same description and you get three gardens, three kitchens, three streets: the words are the same, the picture is not. A longer story needs something the model can see, not only read.

The Scenes node makes the empty location once, as a picture, and draws every frame inside it. Every frame sees that same picture, so the gate, the door and the light stay where they were. Describe the location in a sentence or two, without people in it.

For every scene you write what happens and how it ends. That last frame is also the first frame of the next scene, the same file, so one scene runs into the next without a jump.

• Style: how the whole film looks, written once.
• Location: the empty place, made once.
• Opening frame: where scene 1 starts.
• Per scene: what happens, and the last frame.

Put spoken lines in quotes and say how they are said before the line: mutters, worried: "Steady now." The video model speaks the line in the clip itself, in the same pass as the picture.

Keep a line short enough to fit the scene: one sentence for five seconds.

Run scenes makes the location, then the frames, then the clips, and joins them into one video. Look at the frames before the clips run: a frame costs a fraction of a clip, so that is where you change your mind.