Seedance 2.0 Skill OS is a production pipeline written in Python that enables AI-assisted filmmaking through a directing-first methodology. The repository, maintained by emily2040 and last updated August 1, 2026, provides a modular agent-skill package designed to work with Seedance 2.0 and compatible video generation platforms including ByteDance Seedance 2.0, Runway, Dreamina, Jimeng, Doubao, Volcengine Ark, BytePlus ModelArk, and fal. The core philosophy rejects generic "cinematic" descriptors in favor of scene-specific directorial intent, where camera movement, lighting, blocking, performance, and sound all serve a single dramatic purpose.
The directing engine at the repository's center encodes directorial judgment by reading a scene's dramatic function—the turn, point of view, power, and subtext—and derives a coherent setup rather than stacking adjectives. The system ships with 33 worked derivations across genres including product, music video, horror, anime, action, comedy, documentary, high fashion, and science fiction, each shown end to end. This approach ensures that a reveal is lit, framed, blocked, and performed differently than a goodbye, refusing generic answers in favor of specific ones.
The skill package transforms Seedance work into a repeatable assistant workflow with multiple operational modes. It routes vague ideas into creative interviews before premature prompt writing, writes full or compressed prompts for text-to-video, image-to-video, video-to-video, reference-to-video, and edit workflows, and separates reference assets by role including identity, environment, motion, camera rhythm, audio tempo, style, and endpoint. The system maintains source-dated platform facts so API, pricing, region, quota, and model-ID details are not guessed, and plans into model strengths through capability mapping and fidelity allocation before drafting.
Seedance 2.0 Skill OS provides native entry points in six languages: English, Chinese, Japanese, Korean, Spanish, and Russian. Each language receives first-class documentation including language-specific quickstart guides, vocabulary references, and worked examples. Chinese users can begin with character locking and frame-specific guidance; Japanese readers receive documentation on character identity and costume clarity; Korean users access separated prompting for character, camera, lighting, and sound. The repository maintains consistent reference tag formatting across all languages to ensure model compatibility.
The package includes professional filmmaker workflows for treatment-to-shot-list planning, shot contracts, continuity ledgers, color handoff, audio post-production, subtitles and localization, aspect-ratio variants, campaign cutdowns, and delivery with quality control. For longer stories, the system uses a clip-based continuation workflow where each new clip is based on actual generated footage rather than the original prompt, treating the project state as the source of truth and the accepted generated clip as the determinant of what happens next.
Safety handling is built into the system through IP-safe rewrites that clarify benign production context rather than hiding unsafe intent. The skill can rewrite unsafe requests involving celebrities, protected intellectual property, private persons, brands, logos, songs, or voices into safer creative equivalents. Diagnostic capabilities identify failed outputs with concrete repair levers including camera, lighting, motion, reference role, duration, framing, audio, and safety wording. The repository ships validation scripts, evaluation cases, source data, and design checks so maintainers can review changes before release, maintaining the system as an auditable operating system for Seedance work.
