---
license: mit
tags:
  - image-to-3d
  - trellis2
  - gguf
  - qtmesheditor
---

# QtMeshEditor-trellis2-gguf

GGUF weights for the [TRELLIS.2](https://github.com/microsoft/TRELLIS.2) image-to-3D pipeline, as consumed by
[trellis.cpp](https://github.com/pwilkin/trellis.cpp) and by **QtMeshEditor's `trellis2` backend**
(`qtmesh generate3d --backend trellis2`; QtMeshEditor invokes `trellis-cli --dump-post` and performs its own
game-ready processing and PBR texture baking).

Mirror of [`ilintar/trellis2-gguf`](https://huggingface.co/ilintar/trellis2-gguf) (conversions of the original
checkpoints; thank you!). This mirror pins the exact set QtMeshEditor is tested against. Only the 512-pipeline
subset plus the shared decoders is required by QtMeshEditor; the `*_1024` flows enable the higher-resolution cascade.

## Files & licenses

| file | source checkpoint | license |
|---|---|---|
| `ss_flow.gguf`, `shape_flow_512.gguf`, `tex_flow_512.gguf`, `shape_flow_1024.gguf`*, `tex_flow_1024.gguf`* | [`microsoft/TRELLIS.2-4B`](https://huggingface.co/microsoft/TRELLIS.2-4B) | **MIT** |
| `shape_dec.gguf`, `tex_dec.gguf` | `microsoft/TRELLIS.2-4B` | **MIT** |
| `ss_dec.gguf` | [`microsoft/TRELLIS-image-large`](https://huggingface.co/microsoft/TRELLIS-image-large) | **MIT** |
| `dinov3.gguf` | [`timm/vit_large_patch16_dinov3.lvd1689m`](https://huggingface.co/timm/vit_large_patch16_dinov3.lvd1689m) (ungated mirror of Meta DINOv3 ViT-L) | **[DINOv3 License](https://ai.meta.com/resources/models-and-libraries/dinov3-license/)** (Meta; commercial use permitted, redistribution with conditions) |

\* not uploaded yet if absent — the 512 pipeline works without them.

**Built with DINOv3.** The `dinov3.gguf` conditioning encoder is a converted DINOv3 model distributed under
Meta's DINOv3 License, not MIT — see the license link above; the rest of the repository is MIT.

The overall `license: mit` tag reflects the majority of files; `dinov3.gguf` carries its own license as stated.