Replace placeholder examples with real RoboTwin world-model demo inputs 9960799 verified multimodalart HF Staff commited on 26 days ago
World-model mode: hold flow stream fixed at clean latent, denoise only RGB; drop RoboTwin camera prefix ea48d9e verified multimodalart HF Staff commited on 26 days ago
Prepend trained CAMERA_PREFIX to video prompt to match reference (fix garbled output) b883e8f verified multimodalart HF Staff commited on 26 days ago