Recently, the transfer application of diffusion models in super-resolution tasks has faced the problem of decreased fidelity. Due to the inherent random sampling characteristics of diffusion models, direct application in super-resolution tasks can result in generated details deviating from the true distribution of high-resolution images. To address this, we propose ​DeltaDiff, a novel framework that constrains the diffusion process, its essence is to establish a deterministic mapping path between HR and LR, rather than the random noise disturbance process of traditional diffusion models. Theoretical analysis demonstrates a ​25% reduction in diffusion entropy in the residual space compared to pixel-space diffusion, effectively suppressing irrelevant noise interference. The experimental results show that our method surpasses state-of-the-art models and generates results with better fidelity. This work establishes a new ​low-rank constrained paradigm for applying diffusion models to image reconstruction tasks, balancing stochastic generation with structural fidelity. Our code and model are publicly available at https://github.com/continueyang/DeltaDiff .