Hunyuan3D-2 Turbo — draft-speed image-to-3D

The step-distilled turbo checkpoints from the Hunyuan3D-2 family, single-view and multi-view. ≈6–18 s per mesh on an RTX 5090 — fast enough to iterate on a subject interactively.

What this repo is: the dit-v2-0-turbo (single-view) and dit-v2-mv-turbo (multi-view) shape checkpoints — weights only, not a retrain. The samples below were generated with the single-view turbo.

Samples

Single image in, mesh out. Source images were rendered with Z-Image Turbo; the meshes are Blender Workbench clay renders of the raw .glb output.


_{source — `a cute chunky retro robot toy standing upright, rounded metal body, simple friendly face, studio product photograph on a plain white background`}	_{mesh — 0.72 MB `.glb`, generated in 18 s (incl. model load), seed 42}

_{source — `an ornate ceramic teapot with a curved spout and handle, glazed blue and white porcelain, studio product photograph`}	_{mesh — 0.70 MB `.glb`, generated in 6 s, seed 42}

Textured (`img2mesh_textured`)

The same robot through the full chain — turbo shape, then the paint stage — instead of shape alone.


_{`img2mesh` — 0.72 MB, 18 s}	_{`img2mesh_textured` — 3.4 MB with a baked base-colour texture, +67 s for the paint stage}

Turntable

Four views, 90° apart.

Recommended settings

Parameter	Production value	Meaning
`seed`	any	Deterministic per seed
input	one RGB(A) image	Centred subject, plain background (`mv` variant takes several views)
output	`.glb`	Watertight triangle mesh, no texture

Supported modes: img2mesh, txt2mesh (single-view turbo); img2mesh_mv (multi-view turbo); _textured variants via a separate paint stage

Notes and gotchas

This is the draft tier. Meshes land around 0.7 MB where the larger models in this family return 10–38 MB for the same inputs. Silhouette, proportion and major features are captured; fine relief is smoothed away. Use it to iterate on framing and subject, then re-run a heavier model for the final asset.
Two checkpoints ship here. dit-v2-0-turbo takes one image; dit-v2-mv-turbo takes several views of the same subject and is the one to use when you have them — they are not interchangeable at the job level.
Output is untextured. Colour comes from a separate paint pipeline.
Generated meshes come out facing the source image's front, so a fixed camera works across a batch.

Files

Path	Size	Role
`split/fp16/turbo/hunyuan3d-dit-v2-0-turbo/model.fp16.safetensors`	4.6 GB	single-view turbo shape DiT — the samples above
`split/fp16/turbo/hunyuan3d-dit-v2-mv-turbo/model.fp16.safetensors`	4.6 GB	multi-view turbo shape DiT

Each ships with its config.yaml alongside.

Provenance

Upstream base model: tencent/Hunyuan3D-2 turbo checkpoints
This build: redistributed as a split component tree. Weights are not retrained here.
License: Tencent Hunyuan3D community license — refer to upstream for redistribution and commercial-use terms.

Downloads last month: -

Inference Providers NEW

Image-to-3D

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for ChrisColeTech/hunyuan3d-v2-turbo

Base model

tencent/Hunyuan3D-2

Finetuned

(6)

this model


_{source — `a cute chunky retro robot toy standing upright, rounded metal body, simple friendly face, studio product photograph on a plain white background`}	_{mesh — 0.72 MB `.glb`, generated in 18 s (incl. model load), seed 42}

_{source — `an ornate ceramic teapot with a curved spout and handle, glazed blue and white porcelain, studio product photograph`}	_{mesh — 0.70 MB `.glb`, generated in 6 s, seed 42}