Real-ESRGAN

Real-ESRGAN is a local AI upscaler and restoration toolkit for people enlarging images or improving video, with dedicated models for anime illustrations and animation. It builds on ESRGAN and uses models trained entirely on synthetic data to address degraded images. The code is open source under the BSD 3-Clause license.

Its models cover different kinds of material. RealESRGAN_x4plus handles general images, while RealESRGAN_x4plus_anime_6B targets anime artwork and AnimeVideo-v3 targets animated video. The smaller realesr-general-x4v3 model includes adjustable denoising, so users can balance noise reduction against overly smooth results. GFPGAN integration adds face enhancement.

The Python implementation uses PyTorch and BasicSR. It supports transparent, grayscale and 16-bit images, folder processing, and output sizes beyond the models' native enlargement scales through additional resizing. Developers can also train or fine-tune models on their own data.

Portable builds through Real-ESRGAN-ncnn-vulkan run on Windows, Linux and macOS with Intel, AMD or NVIDIA GPUs. They bundle the models and don't require a CUDA or PyTorch environment. Processing runs on your machine; the Replicate, Colab and Hugging Face Spaces demos run online.

The portable implementation has fewer capabilities than the Python version and can produce visible inconsistencies between image tiles. For people who prefer a graphical interface, projects including Upscayl and Waifu2x-Extension-GUI use Real-ESRGAN.