seedance

seedance is a skill for generating videos with ByteDance's Seedance 2.0 model family through the inference.sh `belt` command-line tool. Seedance 2.0 is a unified model that handles text-to-video, image-to-video, first-and-last-frame control, and multimodal reference-to-video with synchronized audio at up to 1080p and 4–15 second durations. The skill packages this so an agent can produce sound-synced AI video from a single shell command without local model infrastructure.

Four variants are documented: Seedance 2.0 (best quality, up to 1080p), Seedance 2.0 Fast (quicker, up to 720p), and two Studio variants that automatically upload reference images to the BytePlus private virtual portrait library for stronger face and character consistency. Generation mode is inferred from the inputs — prompt only for text-to-video, prompt plus image for image-to-video, plus an end image for first/last-frame control, or reference image/video/audio arrays for multimodal guidance. The skill includes a detailed prompt guide (referencing assets by type and index such as Image 1 or Video 1), formulas for video editing, extension, and stitching, and worked examples for product ads, video element replacement, and reference-driven character scenes.

It targets creators and marketers producing social media videos, music videos, product demos, and animated content with sound. Because it relies entirely on the hosted inference.sh/BytePlus service, it is a documented front-end to those platforms. The Studio variants notably upload user-supplied portrait references to a third-party private library, which is a normal feature of the service but worth noting for anyone handling images of real people.