Give it one episode. It watches the whole thing and writes down how it was made — every shot, its framing, its camera, its lines. The story stays. The look and the cast are yours to change: write the style in your own words, put your own faces and locations in it, move it to a different world entirely, and shoot the same episode again.
Only the dialogue and the story are kept: the beats, the spoken lines, the order of the cuts and where the camera goes. Everything you can see — the place, the clothes, the props — is rebuilt for the new look.
Ground truth for names and lines. Without it they are transcribed by ear, which is where wrong names and half-heard dialogue come from.
Give it one and the episode becomes the direction template: its framing, camera and cutting are applied to your new story, and you get a list of exactly what changed.
The episode is cut every few seconds; Seedance renders a whole sequence in one clip. Its cuts get packed up to the ceiling above, and each pack is generated as one clip with real coverage inside it — fewer generations, and it cuts like the original.
Writes a Seedance prompt for every shot using the structure above and your cast — then opens it in Drama Series to generate.
A style change re-renders the same dungeon. This rebuilds it as an office: the locations, what everyone is wearing, and every prop inside every action — a lit candle becomes a desk lamp because both are the light in the room. The beats and the spoken lines are never sent for rewriting, so they can't drift.
Same story, different medium. Swap the look here and the cast in step 3 — the beats and the dialogue do not move.