---
license: cc-by-nc-4.0 # Assuming a Creative Commons license, adjust if needed.
task_categories:
- image-to-image
tags:
- image-editing
- artistic-style-transfer
- few-shot-learning
---

# PhotoDoodle: Learning Artistic Image Editing from Few-Shot Pairwise Data

[Paper](https://arxiv.org/abs/2502.14397) | [Hugging Face Model](https://huggingface.co/nicolaus-huang/PhotoDoodle)


Authors: [Huang Shijie](https://scholar.google.com/citations?user=HmqYYosAAAAJ), [Yiren Song](https://scholar.google.com.hk/citations?user=L2YS0jgAAAAJ), [Yuxuan Zhang](https://xiaojiu-z.github.io/YuxuanZhang.github.io/), [Hailong Guo](https://github.com/logn-2024), Xueyin Wang, [Mike Zheng Shou](https://sites.google.com/view/showlab), [Liu Jiaming](https://scholar.google.com/citations?user=SmL7oMQAAAAJ&hl=en)
[Show Lab](https://sites.google.com/view/showlab), National University of Singapore


<img src='./assets/teaser.png' width='100%' />


## Dataset

This dataset provides paired image data for artistic image editing. Each entry contains:

* **source:** Path to the original image.
* **target:** Path to the artistically edited image.
* **caption:** A description of the edits applied.


The dataset is available on [Hugging Face Datasets](https://huggingface.co/datasets/nicolaus-huang/PhotoDoodle).  See the [dataset README](./data/README.md) for further details (add if a separate README exists).


## Quick Start (Model Usage)

### Configuration
1. **Environment setup:**
   ```bash
   git clone git@github.com:showlab/PhotoDoodle.git
   cd PhotoDoodle
   conda create -n doodle python=3.11.10
   conda activate doodle
   ```
2. **Requirements installation:**
   ```bash
   pip install torch==2.5.1 torchvision==0.20.1 torchaudio==2.5.1 --index-url https://download.pytorch.org/whl/cu124
   pip install --upgrade -r requirements.txt
   ```

### Inference
The provided code integrates the `diffusers` pipeline with the PhotoDoodle model.  You can run inference using the script:

```bash
python inference.py
```

or directly use the pipeline:  (Code example from original README remains here)

```python
from src.pipeline_pe_clone import FluxPipeline
import torch
from PIL import Image

# ... (rest of the inference code from original README)
```

### Model Weights
[Hugging Face Model](https://huggingface.co/nicolaus-huang/PhotoDoodle) contains the following weights:

| Model Name                     | Description                                    | Resolution |
|---------------------------------|------------------------------------------------|------------|
| `pretrain.safetensors`         | Base PhotoDoodle model                         | 768, 768    |
| `sksmonstercalledlulu.safetensors` | Style model for Cartoon monster                  | 768, 512    |
| `sksmagiceffects.safetensors`   | Style model for 3D effects                     | 768, 512    |
| `skspaintingeffects.safetensors` | Style model for Flowing color blocks            | 768, 512    |
| `sksedgeeffect.safetensors`     | Style model for Hand-drawn outline             | 768, 512    |

**(Note: You need to load and fuse the `pretrained` checkpoint to use the style models.)**


### Results
![R-F](./assets/R-F.jpg)

## Citation
```
@misc{huang2025photodoodlelearningartisticimage,
      title={PhotoDoodle: Learning Artistic Image Editing from Few-Shot Pairwise Data}, 
      author={Shijie Huang and Yiren Song and Yuxuan Zhang and Hailong Guo and Xueyin Wang and Mike Zheng Shou and Jiaming Liu},
      year={2025},
      eprint={2502.14397},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2502.14397}, 
}
```