Show the Signal, Hide the Noise: Spectral Forcing for Pixel-Space Diffusion

2026-06-13

AI Summary

This work addresses the inefficiency in pixel-space diffusion models, which are trained on full-band noise despite informative signal being concentrated in low-frequency components, leading to wasted computational resources in high-frequency regions. To remedy this, the authors propose Spectral Forcing—a parameter-free strategy that dynamically truncates the frequency spectrum prior to patch embedding using a time-conditioned 2D-DCT low-pass filter. This approach explicitly introduces a time-evolving low-pass boundary, thereby making the inherently implicit coarse-to-fine generation process explicit. By steering the model to focus on signal-rich low-frequency content, Spectral Forcing significantly improves both FID and Inception Score on ImageNet-256 and enhances performance on DPG-Bench and GenEval metrics in SenseNova-U1 text-to-image generation.

Abstract

Pixel-space diffusion models are trained on full-bandwidth noisy images, yet the useful signal available to the denoiser is strongly frequency dependent. Under rectified-flow diffusion and natural-image power-law spectra, the per-band data-to-noise contour $k^{*}(t) = (1-t)^{-2/α}$ separates a signal-bearing low-frequency region from a noise-dominated high-frequency region at each time $t$. We show that this implicit coarse-to-fine structure is not merely descriptive: it induces a capacity-allocation problem. A standard pixel-space denoiser must discover the moving bandwidth boundary internally and can spend computation on frequency-time regions where the optimal prediction collapses to deterministic baselines rather than data-distribution modeling. To make this boundary explicit, we introduce Spectral Forcing, a parameter-free, time-conditional 2D-DCT low-pass operator applied to the noisy input before the patch embedder. Its cutoff expands monotonically with the diffusion time and becomes the identity at the data endpoint. Through controlled synthetic experiments, we identify the regime in which the operator is beneficial: coarse patch tokenization and data whose high-frequency content is predominantly noise rather than essential signal. On ImageNet-256 with JiT-700M/32, Spectral Forcing consistently improves both FID and Inception Score across different training epochs, demonstrating robust gains throughout training; at finer tokenization, the spectral forcing is still competitive. We further insert the unchanged operator into SenseNova-U1, a unified text-to-image model, where it improves DPG-Bench and GenEval, showing that the input-side spectral prior transfers beyond class-conditional generation. These results suggest a route to capacity-efficient pixel-space diffusion by showing the signal and hiding the noise.

Problem

• pixel-space diffusion
• spectral forcing
• denoising
• frequency-dependent signal
• capacity allocation

Innovation

• Spectral Forcing
• pixel-space diffusion
• frequency-aware denoising
• 2D-DCT low-pass filtering
• capacity-efficient generation