Prompt engineering by way of Hollywood logistics
A modular agent OS that directs Seedance 2.0 video generation with film-production discipline—shot contracts, continuity rules, and retake budgets—instead of vague prompt dumps.
What it does This repository is a structured skill package that turns Seedance 2.0 video generation into a repeatable, auditable workflow for AI assistants. It replaces one-off prompt dumps with modular stages—creative interviews, reference mapping, prompt compilation, copyright safety, continuation logic, and delivery QC—backed by source-dated platform facts and validation scripts.
The interesting bit The depth of professional filmmaking scaffolding is unusual for a model wrapper: it defines role-specific artifacts for directors, DPs, editors, colorists, and sound teams, and forces the agent to draft shot contracts and continuity ledgers before generating a single frame. It also treats the boring parts—IP rewrites, ACES color handoffs, subtitle localization, and attempt budgets—as first-class citizens.
Key highlights
- 28 modular sub-skills covering the full lifecycle from vague idea to delivery packet
- Native-reader paths and cinematic vocabularies in six languages (English, Chinese, Japanese, Korean, Spanish, Russian)
- Source-dated reference library tracking platform APIs, model names, pricing, and regional quotas so the agent never hallucinates current facts
- Professional role-based workflows: shot contracts, continuity ledgers, retake protocols, and color/audio/localization handoffs
- Built-in safety gates that rewrite unsafe IP, celebrity, or brand requests into safer creative equivalents rather than blocking them blindly
Caveats
- This is a workflow and prompt-framework repository, not a video renderer; you still need Seedance 2.0 platform access to actually generate footage
- Operational details are tightly coupled to Seedance-specific platform surfaces (ByteDance, Runway, fal, etc.), limiting portability to other video models
- Continuation logic requires human verification of generated clips because, as the docs note, Seedance may not end exactly where the prompt expected
Verdict Worth bookmarking if you are building AI video agents or working with Seedance 2.0 at scale and need production-grade continuity, safety, and localization scaffolding. Skip it if you are looking for a quick standalone video generator or a model-agnostic prompting guide.