The march of generative AI continues to set new milestones for creative tools. After AI image generators and video generators, 3D visuals are widely seen as the next frontier. And on that front, the Chinese tech giant Tencent has just made a another leap forward.
Hunyuan3D 2.0 appears to be able to generate 3D assets from 2D images with much better quality than its predecessor and many other image-to-3D tools. And it can even animate them. Some are already suggesting that it could revolutionise VFX and game development.
Only this week, we saw an example of the power of Tencent's HunyuanVideo in the form of a viral short that put Keanu Reeves in Severance. Now Tencent's dropped an updated version of its open-source AI 3D generator Hunyuan3D.
Hunyuan3D 2.0 uses two AI models: Hunyuan3D-DiT to generate the 3D assets and Hunyuan3D-Paint to add textures for improved surface detail. Users provide a 2D reference image and the tool makes multiple 2D views object and builds them into a 3D model.
Cameras at specific heights capture the maximum visible area, capturing details that may be missed on other models, particularly at the top and bottom of objects. A guidance system is designed to improve consistency by ensuring every view matches up. Finally, Hunyuan3D-Studio will allow meshes to be edited and animated in a single workspace.
The output appears to be much improved from the previous model if not as good as Microsoft's Trellis. It also appears to be very fast, generating a 3D model in around 25 seconds (or 10 seconds with a smaller version). That could make it a game-changer for ecommerce, VFX previews and game development, where it could allow test versions of characters to be created more quickly. It could also serve for 3D-printable TTRPG assets.
The files are available on Hugging Face and GitHub.
