Tencent has introduced a simplified version of its generative AI Hunyuan World Model 1.0 called “1.0-Lite”. The new model operates faster and consumes approximately 35 percent less video memory than the original version. “1.0-Lite” requires less than 17 gigabytes of graphics memory, allowing it to be used on regular consumer graphics cards.
Like the main version, “1.0-Lite” is open-source and distributed via GitHub, Hugging Face, and the interactive demo sceneTo3D. Users can review the technical description of the model on arxiv.org. The model is designed to create 3D scenes from text or visual prompts, supports panoramic images, and hierarchical 3D reconstruction.
Hunyuan World Model 1.0 allows for separate editing of objects in the scene, such as moving vehicles, trees, or furniture. A separate “sky sphere” is highlighted, which can be used as a source of dynamic lighting for realistic rendering. The model supports two main input types — “text-to-world” and “image-to-world”, and the created scenes can be exported in mesh format for further work in standard 3D environments.
Ready scenes are interactive 360-degree panoramas, where you can look around and move within certain limits. For wider camera movement capabilities or longer 3D videos, an additional Voyager module is required. Developers note that the model is suitable for working with game engines, VR platforms, and other virtual content creation tools.
In addition to Hunyuan World Model 1.0, Tencent has opened access to Hunyuan3D 2.0 for generating textured 3D models, HunyuanVideo for creating videos using AI, and the Hunyuan-A13B language model with dynamic reasoning.
