Short-Drama Video Prompt Assistant

Problem

Short-drama scripts often contain only scenes, characters, and dialogue. When sent directly to a video model, the result can suffer from jumping shots, drifting character identity, and inconsistent style, or even invent actions not present in the script. The short-drama video prompt assistant turns “script → shots → video units → Seedance 2.0 prompts” into a reviewable workflow.

How It Works

• Input check: expects an existing script, defaults to vertical 9:16 short drama; for multi-episode input, it processes only the first episode and later reuses confirmed characters, voice profiles, and base style.
• Stage 1: script analysis and asset list: extracts characters, scenes, and props; designs stable voice profiles for speaking/OS characters using gender, age impression, and voice traits, then asks for confirmation.
• Stage 2: shot breakdown and storyboard design: splits content into Shot 1, Shot 2, etc., based on emotional arcs, the three-shot method, and pacing; merges shots into video units; records shot size, angle, camera movement, composition, lighting, performance, dialogue, and sound, then checks continuity for axis, gaze, light, and action direction.
• Stage 3: prompt generation: generates Episode X-NN style Seedance 2.0 prompts strictly from the confirmed storyboard; references characters, scenes, and props with @asset_name instead of describing appearance; uses a base style + scene-specific style structure and adds fixed plus tailored constraints.

Boundaries

It is for script-based short-drama prompt generation with prepared reference images, not for writing new scripts or inventing character looks. Each video unit usually needs to stay within roughly 4s to 15s; flashbacks, OS, and UI cases follow the documented edge-case rules.