Tencent has officially released the HY WorldPlay model 1.5, which is the first open real-time interactive experience platform in China. This model can quickly generate unique interactive worlds through simple text descriptions or images. Users can control the virtual camera in real time using a keyboard, mouse, or controller, exploring AI-generated scenes as if playing a game.

image.png

The HY WorldPlay model 1.5 has three core capabilities: First, real-time interactive generation. By using its original Context Forcing distillation scheme and streaming inference optimization, the model can generate high-definition video at 24 frames per second. Second, the model has long-range 3D consistency, meaning that when users leave an area and return, the model can accurately "remember" the 3D structure of that area, ensuring scene coherence. Finally, the model provides diverse interactive experiences, suitable for different styles of games or real-world scenarios, and supports multiple view switching and real-time text-triggered events.

The release of the HY WorldPlay model 1.5 also opens source the most comprehensive real-time world model training system in the industry, covering all aspects including data, training, and inference deployment. The Tencent Hunyuan team detailed technical details such as pre-training, continuous training, and reinforcement learning of self-regressive video models in their technical report. This model uses a visual self-regressive task called Next-Frames-Prediction, breaking the contradiction between real-time performance and geometric consistency, offering new possibilities for world modeling.

The HY WorldPlay model 1.5 not only achieves technological breakthroughs but also opens up broader imagination for the future of AI-generated content. In the future, AI will no longer be limited to generating text, images, or videos but will be able to create complete virtual worlds that can be explored and interacted with in real time. This new model has wide applications in areas such as game development, film production, and virtual reality.

  • Online Experience Website: https://3d.hunyuan.tencent.com/sceneTo3D?tab=worldplay

  • Github Link: https://github.com/Tencent-Hunyuan/HY-WorldPlay

  • Hugging Face Link: https://huggingface.co/tencent/HY-WorldPlay

  • Project Homepage: https://3d-models.hunyuan.tencent.com/world/

Key Points:

🌍  Tencent's HY WorldPlay model 1.5 can quickly generate interactive virtual worlds through text or images, supporting real-time exploration.  

🎮  The model has real-time interaction, long-range 3D consistency, and diverse interactive experiences.  

🔧  The newly opened training system provides new possibilities for the future of AI-generated content.