AI Potluck
Back to Gap Map Model components / Image, video, 3D & music generation

TripoSG

VAST
open weights / Overall score: 1.0

TripoSG is VAST's 1.5B-parameter rectified-flow transformer that generates a 3D mesh from a single image, decoding a signed distance field trained on two million curated image-SDF pairs. A distilled TripoSG-scribble variant generates shapes from a sketch and a text prompt for quick prototyping.

The repository's triposg directory carries code derived from Tencent's FlashVDM under Tencent's own community license, beside the MIT root license.

Openness

3 high confidence
3.0
weights
open(TripoSG and TripoSG-scribble download from Hugging Face without a gate)
data
described(the paper and README describe two million curated image-SDF pairs
code
partial(inference scripts only
license
MIT(the weights and the root of the repository

The TripoSG weights are MIT-licensed and download freely, though part of the inference code is derived from Tencent's FlashVDM and carries Tencent's regional license. Only inference code is published and the training data is described but not released.

Adoption

1 high confidence
1.0

Adoption is measured as Hugging Face downloads of the TripoSG and TripoSG-scribble checkpoints.

Capability

1 medium confidence
1.0

TripoSG is not ranked on the image-to-3D arena's main board for rendered quality, where Hunyuan3D sits in the upper half. It appears only on the arena's mesh-topology board, mid-table.

Verified 2026-09-26