Writing a scene an AI video model can actually film

The scriptwriter on UDRAMA writes prompts for every scene, but creators edit the opening and the prologue themselves, and viewers write their own choices. Here is what a filmable scene looks like.

The shape

Seventy to a hundred and twenty words. Camera move, what the protagonist does, who reacts and how, where the shot lands. One action, one reaction, and the moment it lands. It ends on a held image someone can act from: a face deciding, a door opened, a hand still out.

Name people by description, never by role

The video model has no memory. "The queen" is a new queen. "The red-haired woman in the dark green dress and crimson cloak" is the same woman as last scene. The cast sheet exists so every prompt can say exactly this.

Put dialogue in quotes, with the speaker named

The grey-haired man in the black turtleneck says, "You were never going on."

The model records speech, so a line in the prompt is a line in the film. One or two lines, under twelve words each, that turn the scene. Never narration, never someone addressing the camera.

Repeat the light

The look — light source, palette, texture — goes into every prompt. It is what makes the clips cut together.

Play it for real

If someone strikes, we see it land. If someone reaches, we see the other person decide whether to let them. Not a montage. Physical scenes are filmed, not cut away from; everyone in frame is an adult.

What not to write

• Text, captions, logos, subtitles in the image.
• Two light sources.
• "Beautiful", "cinematic", "dramatic" — adjectives doing a verb's job.
• A scene that ends mid-action. End on the image the next scene starts from.

Try it in the studio: the draft gives you a prologue to edit before anything is filmed.