What problem does it solve?
Turning a screenplay into a shot list that a video generation model can actually execute requires reconciling timing, dialogue, camera moves, and reference images by hand, which is error-prone and hard to audit. This Skill converts a script's beat flow into a deterministic, machine-validated storyboard worklist for AI short drama production.
Core Features & Use Cases
- Three-level storyboard structure: Segments (one video generation call, up to 15 seconds) contain 2-5 second cuts, each claiming script beats, with one keyframe per cut pinned to its cut mark.
- Derived H3 prompts: MiniMax H3 alignment instructions and [Shot k] cut times are computed from cut durations and audited verbatim, with dialogue placed word-for-word in <d> blocks.
- 17 scripted quality gates: Beat coverage, cut duration, dialogue fit, style phrase consistency, reference reconciliation, and an optional shot-recipe gate are all checked by code, not by model self-assessment.
- Use Case: Given a script.json from the novel-script skill, run seed to expand beats, cut each episode into segments, validate against all gates, optionally generate keyframes via codex $imagegen using character and scene sheets as references, then export a per-segment H3 production package.
Quick Start
Ask the agent to storyboard episode 1 of your short drama by running the novel-storyboard seed and validate commands against your script.json, then render the HTML review report.
