muapi-seedance-2

Seedance 2.0 Cinema Expert is a media-generation orchestration skill for driving ByteDance's Seedance 2.0 video model (accessed via MuAPI) with director-level, instructional prompting. It frames Seedance 2.0 as an instructional rather than descriptive model and teaches technical cinematography, physics directives, and precise camera grammar to produce high-fidelity video. It solves the problem of getting consistent, controllable output from a complex multimodal video model by codifying its modes, tiers, input limits, and prompt structure.

The skill documents nine core competencies: text-to-video, image-to-video, video extension, first-and-last-frame interpolation, omni reference (images plus audio plus character references), omni-reference training for persistent characters, character sheets, video editing, and watermark removal. It details three tiers (Chinese, Global, VIP) with differing censorship levels, aspect ratios, durations, and quality parameters, plus a fast-queue flag; input limits for images, videos, and audio; and restrictions such as disallowing realistic human faces in uploads except in character/training modes. It defines an '@' reference system for assigning roles to uploaded assets, a six-component Director Brief structure, time-segmented prompting for longer clips, negative prompting, and camera-language references. One documented mode strips Seedance 2.0 watermarks from generated video.

It is aimed at creators, marketers, and video producers generating cinematic AI video, animating reference images, or building identity-consistent characters. Two behaviors warrant noting: it invokes an external third-party generation API (MuAPI), and it includes a watermark-removal capability that removes provenance markings from AI-generated video, which is a content-integrity and terms-of-service concern rather than a system-level security risk.