DDNM solves image restoration problems by leveraging the null-space of the degradation operator combined with pre-trained denoising diffusion models. Users provide a degraded image and specify the restoration task (e.g., denoising, super-resolution), and the method iteratively refines the result without requiring any training on specific degradation pairs. The approach works across arbitrary image sizes and supports applications including old photo restoration, compressed sensing, and blind inpainting.