amd/realesrgan-x4plus-anime-6b

Real-ESRGAN x4plus Anime 6B

This repository hosts the RealESRGAN_x4plus_anime_6B.pth pre-trained generator weights from the xinntao/Real-ESRGAN project. The file is a 1:1 mirror of the asset originally released by Xintao Wang on August 31, 2021 as part of Real-ESRGAN v0.2.2.4.

Per the v0.2.2.4 release notes, this checkpoint is "optimized for anime images with much smaller model size." It is a 6-block RRDBNet — significantly smaller (18 MB) than the standard 23-block x4plus model (67 MB) — and is intended for upscaling anime / illustration content. Visual comparisons with waifu2x are documented in the upstream docs/anime_model.md.

📋 Model Details

Field	Value
Original release	v0.2.2.4, 31 Aug 2021
Authors	Xintao Wang, Liangbin Xie, Chao Dong, Ying Shan — Tencent ARC Lab; Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences (paper)
Architecture	RRDBNet(num_in_ch=3, num_out_ch=3, num_feat=64, num_block=6, num_grow_ch=32, scale=4) (source). The _6B suffix denotes 6 RRDB blocks, vs. 23 in the standard x4plus.
Upscale factor	4×
Weight file	RealESRGAN_x4plus_anime_6B.pth (~18 MB)
Domain	Anime / illustration imagery
Paper	Wang et al., 2021 — Real-ESRGAN: Training Real-World Blind Super-Resolution with Pure Synthetic Data (ICCVW 2021)
License	BSD 3-Clause, Copyright (c) 2021 Xintao Wang
Source repository	github.com/xinntao/Real-ESRGAN
Original asset URL	github.com/xinntao/Real-ESRGAN/releases/download/v0.2.2.4/RealESRGAN_x4plus_anime_6B.pth
Reference doc	docs/anime_model.md

⚡ Intended Use

This model is purpose-built for upscaling anime / line-art / illustration content at 4×. Its compact 6-block architecture makes it roughly 4× smaller than the standard x4plus checkpoint and correspondingly faster at inference. For natural photographs and general scenes the larger amd/realesrgan-x4plus (23-block) checkpoint typically gives better results.

🛠️ How to Use

The canonical entry point is the upstream Real-ESRGAN repository. The workflow below mirrors the PyTorch Inference section of the upstream anime_model.md:

# 1. Clone Real-ESRGAN
git clone https://github.com/xinntao/Real-ESRGAN.git
cd Real-ESRGAN

# 2. Install dependencies
pip install basicsr facexlib gfpgan
pip install -r requirements.txt
python setup.py develop

# 3. Download the weights from this Hugging Face repo
huggingface-cli download amd/realesrgan-x4plus-anime-6b RealESRGAN_x4plus_anime_6B.pth --local-dir weights

# 4. Run inference
python inference_realesrgan.py -n RealESRGAN_x4plus_anime_6B -i inputs

A portable NCNN executable variant is also available via the realesrgan-x4plus-anime model name in Real-ESRGAN-ncnn-vulkan — see the upstream README.md for the full set of options.

⚠️ Caveats and Recommendations

This is a domain-specific model tuned for anime / illustration content. Quality on natural photos will typically be lower than with the standard 23-block x4plus checkpoint. Wang et al. (2021) generally note that Real-ESRGAN can still introduce aliasing or artifacts on difficult inputs; see the upstream anime_model.md for qualitative comparisons against waifu2x.

📌 Citation

If you use this model, please cite the original Real-ESRGAN paper:

@InProceedings{wang2021realesrgan,
    author    = {Xintao Wang and Liangbin Xie and Chao Dong and Ying Shan},
    title     = {Real-ESRGAN: Training Real-World Blind Super-Resolution with Pure Synthetic Data},
    booktitle = {International Conference on Computer Vision Workshops (ICCVW)},
    date      = {2021}
}

📜 License

These weights are distributed under the BSD 3-Clause License, Copyright (c) 2021 Xintao Wang (upstream LICENSE). This repository re-hosts the original artifact unchanged; please attribute the original authors when using or redistributing the weights.

🤗 Acknowledgments

All credit for the model architecture, training methodology, and weights goes to Xintao Wang and the Real-ESRGAN authors at Tencent ARC Lab and the Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences. This Hugging Face repository exists only as a convenient mirror of the pre-trained weight file alongside its license and citation context.