Zhiyuan releases multimodal world model Wujie · Emu3.5, which can achieve cross scene embodied operation

On October 30th, Zhiyuan released the multimodal world model Emu3.5, which achieved "Next State Prediction (NSP)" for multimodal sequences through autoregression and obtained the ability to generalize world modeling. At the scene application level, the model can not only achieve cross scene embodied operations, generalized action planning, and complex interaction capabilities, but also complete text and image generation, image editing, and spatiotemporal transformation.