Why this matters
Most prompt toolkits ask for a vague “cinematic” look; this package asks what the scene is doing and compiles one coherent directorial instruction. It is an agent-focused skill OS that reads a scene’s dramatic function, picks a single intention, and produces production-ready prompts and contracts instead of adjective lists. That intent-first approach is designed to keep identity, motion, camera, lighting, and sound aligned across every clip of a longer story.
What Sets It Apart
- Intent-first directing engine: reads scene beats and derives a single directorial voice rather than stacking generic quality words. This yields compact, actionable prompts (T2V, I2V, V2V, R2V, FLF2V, edit/extend, audio-aware, first/last-frame workflows).
- Multimodal reference and role separation: enforces explicit roles for references (identity, environment, motion, camera rhythm, audio tempo, style, endpoint) so assets are used predictably across generations.
- Production-grade workflows: shot contracts, continuity ledgers, ACES color handoff notes, audio stems guidance, localization/subtitle guidance, delivery/QC checklists, and a five-verdict retake protocol for iterative generation.
- Safety and governance: built-in rewrites for celebrity/IP/brand/voice requests, false-positive repair strategies (clarify context instead of hiding intent), and source-dated platform claims to avoid stale API assertions.
- Multilingual and community-informed: native reader entry points and prompt vocab for English, 中文, 日本語, 한국어, Español, and Русский; the v6 line ships 33 worked genre derivations and localized prompt examples.
- Platform-aware limits: documents that public sources (as of mid‑2026) describe Seedance 2.0 as accepting text, images, audio, and video references (up to 9 images, 3 video clips, 3 audio clips) and tracks model/surface IDs per provider.
Who it’s for and tradeoffs
Great fit if you are a filmmaker, agency producer, or agent developer who needs reproducible, multi-clip directed outputs rather than one-off “make it cinematic” prompts. The package is especially useful when you must preserve continuity across multiple generations, separate reference roles, or hand off artifacts to post teams.
Look elsewhere if you only need casual, single-shot prompt templates or a lightweight GUI: this skill OS is written as a professional agent-skill package (installable into agent clients) and assumes a production workflow mindset. Also, platform-specific claims (API endpoints, pricing, face/portrait rules, upload limits) are intentionally source-dated and must be rechecked before implementation for each provider.
Practical notes
The repository is distributed as a modular skill set with install scripts and CI validation checks; it keeps dense facts in a reference library and emphasizes validation runs for evals and continuity. Current release line in the README is v6.7.0 and the project is MIT-licensed.
