hiramdeng/refineart-model-mirror

RefineArt Model Mirror

This repository serves as a model mirror for the RefineArt project, providing reliable access to the large model files required for 4K image enhancement using APISR and SUPIR pipelines.

Model Catalog

Model File	Type	Purpose	Size	Format
SUPIR-v0Q_fp16.safetensors	SUPIR	Photo-realistic restoration (Quality variant)	~4.2 GB	FP16
SUPIR-v0F_fp16.safetensors	SUPIR	Photo-realistic restoration (Fidelity variant)	~4.2 GB	FP16
SUPIR-v0Q.ckpt	SUPIR	Photo-realistic restoration (Quality variant)	~8.4 GB	Full
SUPIR-v0F.ckpt	SUPIR	Photo-realistic restoration (Fidelity variant)	~8.4 GB	Full
sd_xl_base_1.0_0.9vae.safetensors	Base Model	Required SDXL base for SUPIR	~6.9 GB	FP16

APISR models are included in the main RefineArt repository under src/magine/libs/apisr/weights/.

Overview

RefineArt integrates two state-of-the-art super-resolution models:

1. SUPIR (Scaling Up to Excellence for Photo-Realistic Image Restoration) - Designed for realistic photo enhancement
2. APISR (Anime Picture Intelligent Super Resolution) - Specialized for anime and illustrations

Due to the large size of these model files and potential availability issues with the original sources, this repository acts as a reliable mirror to ensure smooth operation of the RefineArt application.

SUPIR Models

About SUPIR

SUPIR (Scaling Up to Excellence for Photo-Realistic Image Restoration) originates from the CVPR 2024 paper:

Fanghua Yu, Jinjin Gu, Zheyuan Li, Jinfan Hu, Xiangtao Kong, Xintao Wang, Jingwen He, Yu Qiao, Chao Dong. "Scaling Up to Excellence: Practicing Model Scaling for Photo-Realistic Image Restoration In the Wild." CVPR 2024.

Variants

• SUPIR-v0Q: Quality variant optimized for visual quality
• SUPIR-v0F: Fidelity variant optimized for preserving original details
• FP16 versions: Memory-efficient half-precision variants
• Full versions: Complete precision models

Required Dependencies

• sd_xl_base_1.0_0.9vae.safetensors: SDXL base model required for SUPIR operation

Sources

• Original Research: Project Page | Paper
• Official Repository: Fanghua-Yu/SUPIR
• Community Mirror: Kijai/SUPIR_pruned

APISR Models

About APISR

APISR (Anime Picture Intelligent Super Resolution) is specifically designed for enhancing anime and illustration artwork. It incorporates components from proven super-resolution frameworks.

Model Types

• Real-ESRGAN Checkpoints: Various models optimized for different anime enhancement scenarios
• RealCUGAN Models: Multiple variants with different upscaling factors and denoising levels

Sources

• Original Implementation: Kiteretsu77/APISR
• Core Components:
  • Real-ESRGAN
  • BasicSR

Usage

Automatic Download (Recommended)

The RefineArt project automatically downloads required models from this mirror during first execution:

1. When you run RefineArt for the first time, it will automatically download the required SUPIR models to: refineart/src/magine/libs/supir/checkpoints/

2. APISR models are included directly in the main repository at: refineart/src/magine/libs/apisr/weights/

Manual Download

To manually download models for offline use:

1. Clone this repository:

   git clone https://huggingface.co/dzhsurf/refineart-model-mirror
   

2. Copy the required model files to the appropriate directories in your RefineArt installation:

   • SUPIR models: refineart/src/magine/libs/supir/checkpoints/
   • APISR models: Already included in refineart/src/magine/libs/apisr/weights/

License

The model files in this repository are subject to their respective original licenses:

SUPIR License

The SUPIR models are provided for non-commercial use only as per the original project's declaration:

The SUPIR ("Software") is made available for use, reproduction, and distribution strictly for non-commercial purposes. For the purposes of this declaration, "non-commercial" is defined as not primarily intended for or directed towards commercial advantage or monetary compensation.

For commercial use inquiries, please contact the original authors.

APISR and Related Components

• Real-ESRGAN: BSD 3-Clause License
• BasicSR: Apache 2.0 License
• APISR: MIT License

Please ensure compliance with all applicable licenses when using these models.

Attribution

We gratefully acknowledge the original researchers and developers who created these amazing models:

SUPIR Team

Fanghua Yu, Jinjin Gu, Zheyuan Li, Jinfan Hu, Xiangtao Kong, Xintao Wang, Jingwen He, Yu Qiao, Chao Dong

APISR and Component Teams

• APISR Team: Kiteretsu77 and contributors
• Real-ESRGAN Team: Xintao Wang and contributors
• BasicSR Team: XPixelGroup and contributors

Contributing

This repository is maintained as a mirror for the RefineArt project. If you encounter any issues with the model files or need to update them, please open an issue in the RefineArt repository.