Hunyuan3D-2 mini

Image in, mesh out. Hunyuan3D-2 mini is a 0.6B flow-matching diffusion transformer from Tencent that generates the shape of the object in a photo as a watertight mesh, exported as .glb for Blender, Godot, Unity or a browser.

Run it

The mesh lands next to the image as chair-hunyuan3d-2mini.glb. The first run downloads the weights.

Faster: the turbo variant

Step-distilled with FlashVDM: 5 steps instead of 30. Seconds rather than a minute, at a small cost in detail.

From text

The prompt is drawn as an image first (HunyuanDiT), then meshed. A photo you picked is usually better: the mesh reproduces whatever the image gets wrong.

Tips

• One whole object, centred, on a plain background. Backgrounds are removed automatically, but a clean input still wins.
• --octree 380 for finer geometry, --faces 20000 for a lighter mesh.
• Shape only needs ~5 GB of VRAM. --texture adds the paint model (~14 GB) and needs an NVIDIA card with the CUDA toolkit to build its extensions.

Licensed under the Tencent Hunyuan 3D Community License — read it before commercial use, and note its territory restrictions.