During training, diffusion models learn a perfect relationship between noise level (SNR) and denoising step (T). During inference, this relationship breaks as the model's own predictions introduce errors, creating SNR values it never trained on for a given step. This causes compounding errors and quality loss.
The SNR-T bias can be fixed efficiently without retraining models. At each denoising step, the image is broken into frequency bands using wavelets. Each band is then given a small correction based on its specific noise mismatch before being recombined. This surgical approach is computationally cheap and universally effective.
Diffusion models naturally reconstruct images in layers. In early denoising stages with high noise, they focus on low-frequency information like overall composition and color. As noise decreases in later steps, they add high-frequency details like textures and sharp edges. This hierarchical process is key to understanding their behavior.