On September 10, Xiaomi Technology unveiled the open-source release of the Xiaomi-Robotics-U0 embodied world model training framework, along with its model weights. This comprehensive open-source package encompasses model weights in two distinct scales: 4B and 34B. These weights are designed to support a wide array of tasks, including embodied scene generation and editing, single-image and multi-image generation, video generation, as well as the generation of text-image interleaved sequences. Additionally, it offers a corresponding training and inference framework, with the primary goal of fostering advancements in world model research, embodied intelligence, and robotics applications. By open-sourcing these models and their underlying code, Xiaomi aims to reduce technical barriers and expedite the integration and deployment of world model technologies in real-world robotic environments.