A 722M-parameter image-to-video model distilled from Causal Forcing / Wan2.1 (model).
It keeps your picture and starts the motion you describe. The second half of the clip tends to melt, and that's part of the charm.
Images are center-cropped to landscape 832×480.
