DiffBIR is an open-source project that utilizes generative adversarial networks for blind image restoration. It is capable of denoising, super-resolution, and color correction on low-quality images without the need for corresponding high-quality images. The project employs a two-stage network architecture: the first stage uses a model to remove low-level degradations such as coded blur and JPEG compression noise, while the second stage leverages a pre-trained model to guide the network in generating high-quality images. DiffBIR's models are compact and efficient, making them deployable on a variety of devices. It also offers an easy-to-use command line interface and Web Demo, allowing users to experience its powerful restoration capabilities.