---
base_model: microsoft/TRELLIS-image-large
pipeline_tag: image-to-3d
license: cc-by-nc-4.0
tags:
- image-to-3d
- 3d
- stylization
- trellis
---

# DreamStyle3D

DreamStyle3D is an efficient framework for generating stylized 3D assets from a content image
and a style image, presented in the paper [DreamStyle3D: Efficient 3D Stylized Asset Generation via Dual-Attention Disentanglement](https://huggingface.co/papers/2607.24721).

## Files

The model repository contains the model configuration and safetensors weights:

```text
dreamstyle3d_slat_flow.json
dreamstyle3d_slat_flow.safetensors
```

## Usage

From the DreamStyle3D source repository, download TRELLIS and install this checkpoint with:

```bash
python scripts/install_models.py \
  --dreamstyle-repo jiangmu/DreamStyle3D
```

The installer places both files under
`ckpts/trellis-image-large/ckpts/dreamstyle3d/` and updates the TRELLIS `pipeline.json` entry to:

```json
"slat_flow_model": "ckpts/dreamstyle3d/dreamstyle3d_slat_flow"
```

## SHA-256

```text
d581edb3e57ef92dbe553651b9576ba1f536c274cb4d0e74d1d610513005103f  dreamstyle3d_slat_flow.json
fd0e00fa691df80cb03befbec1f68bcf1f03d1323db367b0d129a4eb84921ee6  dreamstyle3d_slat_flow.safetensors
```