README.md

---
library_name: transformers
license: apache-2.0
pipeline_tag: image-to-image
tags:
- multimodal
- image-generation
- image-to-image
- masked-diffusion
---

# Refined UniDFlow

This repository contains the released UniDFlow weights, tokenizer, and VQ-VAE
prepared for the modular `unidflow` Python package.

```python
from unidflow import UniDFlowPipeline

pipe = UniDFlowPipeline.from_pretrained("onkarsus13/unidflow-base-7B")
result = pipe(
    "image-to-image",
    image="input.jpg",
    prompt="Turn this photograph into a watercolor painting",
    edit_type="edit",
    seed=42,
)
result.save("edited.png")
```

UniDFlow is a bidirectional masked-diffusion model. Use `UniDFlowPipeline` rather
than the autoregressive `transformers.generate()` method. Preserve the upstream
UniDFlow citation and notices when redistributing this checkpoint.