One-Step Residual Shifting Diffusion for Image Super-Resolution via Distillation

Diffusion models for super-resolution (SR) produce high-quality visual results but require expensive computational costs. Despite the development of several methods to accelerate diffusion-based SR models, some (e.g., SinSR) fail to produce realistic perceptual details, while others (e.g., OSEDiff) may hallucinate non-existent structures. To overcome these issues, we present RSD, a new distillation method for ResShift. Our method is based on training the student network to produce images such that a new fake ResShift model trained on them will coincide with the teacher model. RSD achieves single-step restoration and outperforms the teacher by a noticeable margin in various perceptual metrics (LPIPS, CLIPIQA, MUSIQ). We show that our distillation method can surpass SinSR, the other distillation-based method for ResShift, making it on par with state-of-the-art diffusion SR distillation methods with limited computational costs in terms of perceptual quality. Compared to SR methods based on pre-trained text-to-image models, RSD produces competitive perceptual quality and requires fewer parameters, GPU memory, and training cost. We provide experimental results on various real-world and synthetic datasets, including RealSR, RealSet65, DRealSR, ImageNet, and DIV2K. We provide the code at https://github.com/Daniil-Selikhanovych/RSD.

Enhancing blurry or pixelated photos to high quality using AI can produce stunning results, but the most powerful methods are slow and demand significant computing power. Faster alternatives often compromise: some generate bland, lifeless outputs, while others invent details that never existed in the original scene. We introduce RSD, a new technique that efficiently transfers expertise from a powerful but slow AI model into a fast, single-step version. Instead of simply copying results, our method trains the fast model to produce images that would lead the original model to make identical decisions - preserving visual fidelity while accelerating processing. RSD delivers sharper, more realistic results than previous fast methods, as confirmed by both technical measurements and visual results. Because it requires less computing power, it can run on more accessible hardware and reduces energy consumption. We release our code openly, enabling researchers and developers to apply efficient, high-quality image enhancement in fields ranging from photography to medical imaging and satellite analysis.