---
title: UltraTex
canonical_url: "https://www.modelscope.ai/models/Yibo-Zhang/UltraTex"
md_url: "https://www.modelscope.ai/models/Yibo-Zhang/UltraTex.md"
repository: Yibo-Zhang/UltraTex
last_updated: 2026-10-03
license: mit
pipeline_tag: text-generation
tasks:
  - text-generation
base_model_relation: finetune
parameters: 358.6M
tensor_type:
  - F32
  - BF16
library_name:
  - safetensors
  - pytorch
frameworks:
  - pytorch
downloads: 156
stars: 0
tags:
  - texture-generation
  - 3d
  - flux
---

# UltraTex

> UltraTex - An open-source model by Yibo-Zhang on ModelScope. UltraTex weights for multi-view texture generation at 2048 resolution. Code: https://github.com/yiboz2001/UltraTex

Yibo-Zhang/UltraTex is a 358.6M-parameter text-generation model on ModelScope. licensed under mit.

- **Repository**: Yibo-Zhang/UltraTex
- **License**: mit
- **Tasks**: text-generation
- **Parameters**: 358.6M
- **Tags**: texture-generation, 3d, flux
- **Downloads**: 156
- **Stars**: 0
- **Last updated**: 2026-10-03

Source: https://www.modelscope.ai/models/Yibo-Zhang/UltraTex

---

# UltraTex

UltraTex weights for multi-view texture generation at 2048 resolution.
Code: https://github.com/yiboz2001/UltraTex

Each backbone has its own LoRA and its own foreground-aware VAE decoder; they are
not interchangeable.

| File | Backbone | Purpose |
|---|---|---|
| `flux1/lora/dit_lora.safetensors` | FLUX.1-dev | albedo LoRA |
| `flux1/decoder.pt` | FLUX.1-dev | foreground-aware VAE decoder |
| `flux2/lora/dit_lora.safetensors` | FLUX.2-Klein base 9B | albedo LoRA |
| `flux2/decoder.pt` | FLUX.2-Klein base 9B | foreground-aware VAE decoder |
| `flux2_mr/lora/dit_lora.safetensors` | FLUX.2-Klein base 9B | metallic-roughness LoRA (uses `flux2/decoder.pt`) |

## Usage

Download into the repository's `checkpoints/` directory, keeping the layout above:

```bash
modelscope download --model Yibo-Zhang/UltraTex --local_dir checkpoints
```

Base models (FLUX.1-dev, FLUX.2-Klein base 9B, Qwen3-8B, T5, CLIP) come from their
original sources; see the repository README for the expected `checkpoints/base/` layout.

```bash
bash scripts/inference_flux1.sh
bash scripts/inference_flux2.sh
bash scripts/inference_flux2.sh standard --task mr
```