pkuliyi2015/multidiffusion-upscaler-for-automatic1111
An extension for Automatic1111's Stable Diffusion web UI that enables generating and upscaling large images (≥2K) with limited VRAM (≤6GB) using advanced tiled diffusion and VAE methods.
Core Features
Detailed Introduction
This project is a crucial extension for users of the Automatic1111 Stable Diffusion web UI, addressing the common challenge of generating or upscaling high-resolution images with constrained GPU memory. By implementing state-of-the-art tiled diffusion techniques, including Mixture of Diffusers, MultiDiffusion, Demofusion, and original Tiled VAE/Noise Inversion methods, it allows users to produce large-scale artworks (2K and above) even with as little as 6GB of VRAM. It significantly enhances the accessibility and capability of AI image generation for a broader range of hardware configurations.