StableSR
StableSR is a state-of-the-art real-world image super-resolution model that leverages diffusion priors to recover high-quality details from low-resolution inputs. Developed by researchers at Nanyang Technological University and published in the International Journal of Computer Vision, the software utilizes an effective method to exploit conditional attention and diffusion-based generation for superior restoration. It excels in handling complex real-world degradations such as noise, blur, and compression artifacts, producing natural and realistic textures rather than smooth averages. The project supports integration with popular tools including WebUI, ComfyUI, and ModelScope, and offers compatibility with SD-Turbo for accelerated inference. Features include support for DDIM sampling, negative prompts, and specialized training scripts for face super-resolution. The software is available as an open-source repository with demonstrations hosted on Hugging Face, Replicate, and OpenXLab. It is designed for research