Upscale-Image

<div align="center">

RealRestorer: Towards Generalizable Real-World Image Restoration with Large-Scale Image Editing Models

</div>

RealRestorer is presented in the paper RealRestorer: Towards Generalizable Real-World Image Restoration with Large-Scale Image Editing Models.

RealRestorer is a generalizable real-world image restoration model built on top of large-scale image editing models. It is designed to restore degraded real images while preserving the original scene structure, semantic content, and fine-grained details.

🔥🔥🔥 News!!

• Mar 2026: 👋 We release RealRestorer, its checkpoints, and RealIR-Bench.

⚡️ Model Usage

Installation

This project currently relies on the patched local diffusers/ checkout from the RealRestorer repository.

git clone https://github.com/yfyang007/RealRestorer.git
cd RealRestorer

python3.12 -m pip install --upgrade pip

cd diffusers
python -m pip install -e .
cd ..

python -m pip install -r requirements.txt
python -m pip install -e .
python -m pip install -r RealIR-Bench/requirements.txt

You can verify the environment with:

python -c "from diffusers import RealRestorerPipeline; print(RealRestorerPipeline.__name__)"

Inference with Diffusers

import torch
from PIL import Image
from diffusers import RealRestorerPipeline

pipe = RealRestorerPipeline.from_pretrained(
    "RealRestorer/RealRestorer",
    torch_dtype=torch.bfloat16,
)
pipe.enable_model_cpu_offload()

image = Image.open("examples/input.png").convert("RGB")
prompt = "Please deblur the image and make it sharper"

result = pipe(
    image=image,
    prompt=prompt,
    negative_prompt="",
    num_inference_steps=28,
    guidance_scale=3.0,
    seed=42,
    size_level=1024,
)

result.images[0].save("realrestorer_output.png")

CLI Quick Start

python3 infer_realrestorer.py \
  --model_path /path/to/realrestorer_bundle \
  --image /path/to/input.png \
  --prompt "Restore the details and keep the original composition." \
  --output /path/to/output.png \
  --device cuda \
  --torch_dtype bfloat16 \
  --num_inference_steps 28 \
  --guidance_scale 3.0 \
  --seed 42

Recommended Inference Config

• Device: cuda
• Torch dtype: bfloat16
• Inference steps: 28
• Guidance scale: 3.0
• Recommended seed: 42

Example Prompts

Additional Resources

• Project Page: RealRestorer Project Page
• GitHub Repository: yfyang007/RealRestorer
• Model: RealRestorer/RealRestorer
• Degradation Models: RealRestorer/RealRestorer_degradation_models
• Benchmark: RealRestorer/RealIR-Bench

License and Disclaimer

The code of RealRestorer is intended to be released under the Apache License 2.0, while the RealRestorer model and associated benchmark assets are intended for non-commercial academic research use only.

All underlying base models and third-party components remain governed by their original licenses and terms. Users must comply with all applicable upstream licenses when using this project.

Citation

If you find RealRestorer useful in your research, please star and cite:

@misc{yang2026realrestorergeneralizablerealworldimage,
      title={RealRestorer: Towards Generalizable Real-World Image Restoration with Large-Scale Image Editing Models},
      author={Yufeng Yang and Xianfang Zeng and Zhangqi Jiang and Fukun Yin and Jianzhuang Liu and Wei Cheng and jinghong lan and Shiyu Liu and Yuqi Peng and Gang YU and Shifeng Chen},
      year={2026},
      eprint={2603.25502},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2603.25502},
}