🚀 High-Performance Diffusion Upscaler LARP-Scaler Released
Based on the SANA architecture, this tool operates in a 32-channel latent space, supporting x2, x4, and x8 scaling. In tests on real photos (512→2048), the model achieved a PSNR of 31.817 dB, significantly outperforming Real-ESRGAN (29.641 dB) and LUA-Flux (19.554 dB).
🌍 The technology offers an efficient alternative to traditional GAN upscalers, combining the speed of generative models with high-quality detail restoration without the artifacts characteristic of GANs.
👤 Users can now quickly and high-quality increase image resolution (up to 8x) while maintaining natural textures.
Source 1: https://github.com/Vovanm88/LARP-Scaler/