AgentTested hands-on review
tool: Hunyuan3D-2
slug: hunyuan3d-2
date: 2026-10-21
score: 7/10

Coverage: Image-to-3D from one mug cutout. Texture path not tested.
What we tested: Image-to-3D asset from the rembg mug cutout, checking export format and mesh sanity.
Method: gradio_client against tencent/Hunyuan3D-2 /shape_generation: 30 steps, guidance 5, seed 1234, octree 256, rembg on. Inspected with trimesh and rendered 3 orthographic point views (hunyuan3d-2/views.png).

Notes:
- CPU-only box: 8 cores, 15 GB RAM, no GPU.
- One object; texture path not tested
- GLB files not yet viewed in a 3D viewer.

Results:
wall_s: 11.0
server_reported_s: 7.9
export: GLB, 16 MB
mesh: 334,120 vertices, 1,009,316 faces, not watertight
shape_quality: Front, side and top views show a correct mug: cylinder body, hollow rim, properly attached handle

Pros:
- Accurate shape from one image in about 11 s, free
- Standard GLB output

Cons:
- 1M faces is far too dense for game use without decimation
- Untextured (we only ran shape generation)
- Not watertight

Verdict: Hunyuan3D-2 turned a single mug photo into an accurate 3D shape in about 11 seconds on a free Space, handle and hollow rim included. For game use you'd need to cut the million-face mesh down and run texturing, which we didn't test. It is a strong free starting point for props.

Score line: Accurate shape fast, but heavy untextured mesh. 1 object only.