FlowWAM — Optical Flow as a Unified Action Representation
A dual-stream video diffusion model (built on Wan2.2-TI2V-5B) run in world-model mode: the optical-flow stream is held fixed as a clean motion condition while the model denoises a future RGB video from one image and a short text instruction. From the paper FlowWAM: Optical Flow as a Unified Action Representation for World Action Models.
Give it a starting frame and describe the motion — it imagines how the scene evolves under the flow condition.
Examples