SUPIR is a groundbreaking image repair method utilizing the power of generative priors and model expansion. Through multi-modal techniques and advanced generative priors, SUPIR has made significant strides in intelligent and realistic image repair. Model expansion, a key catalyst within SUPIR, has significantly enhanced its capabilities, revealing new potential in image repair. We trained our model on a dataset of 20 million high-resolution, high-quality images, each accompanied by descriptive textual annotations. SUPIR can repair images based on text prompts, broadening its application range and potential. Furthermore, we introduced negative quality prompts to further improve perceived quality. We also developed a repair-guided sampling method to mitigate the fidelity issues encountered in generative repair. Experiments have demonstrated SUPIR's outstanding repair effects and its novel ability to manipulate repairs through text prompts.