A 722M-parameter image-to-video model distilled from Causal Forcing / Wan2.1 (model).
It keeps your picture and starts the motion you describe; past 2 seconds it keeps going in 1-second windows and
gets steadily meltier, which is part of the charm. Images are center-cropped to landscape 832×480.
Sound is generated by MMAudio (weights CC-BY-NC 4.0: non-commercial use).
