Wan Dancer: Music to Dance is a ComfyUI workflow that turns a music clip and a single reference image into a coherent, short dance video. It uses the Wan Dancer model inside a Subgraph (node ID f7467834-35a6-42fe-b525-7f17383beb4f) to generate all frames in one pass, preserving global structure and temporal continuity. You provide a portrait or full-body image in LoadImage and an audio file in LoadAudio; the subgraph handles conditioning and sampling, and SaveVideo assembles the output clip.

Technically, the subgraph packs the full Wan Dancer pipeline: it encodes your reference image as appearance/style conditioning, derives audio conditioning from the waveform, and samples a fixed-length video (149 frames by default) with long-range temporal attention so poses and motion stay consistent across time. The default configuration produces a ~5 second clip. Because the model samples many frames at once, it is VRAM-intensive and benefits from a high-memory GPU. SaveVideo then writes the frames to a video file at an appropriate frame rate for smooth playback.