DiFA: Inference-Time Forward-Process Alignment for Diffusion Models
2026-07-20 • Machine Learning
Machine Learning
AI summaryⓘ
The authors show that current diffusion models treat image generation like a math problem, ignoring natural randomness in the process. They propose a new method called DiFA that looks at the generation steps as connected observations, similar to how a Kalman filter works. This approach helps combine past predictions consistently and reduces blurry results by keeping important details. Tested on popular image datasets, DiFA improved the quality of generated images without extra training. Overall, their method better accounts for the uncertainty inherent in generating images with diffusion models.
Diffusion modelsNumerical integrationDenoising processKalman filteringSequential state estimationTemporal consensusImage generationFID scoreImageNetCIFAR-10
Authors
Shigui Li, Delu Zeng
Abstract
The prevailing inference framework for diffusion models formulates generation fundamentally as a problem of numerical integration. This perspective casts the model as an exact estimator, neglecting the inherent statistical uncertainty of the denoising process. In this work, we propose Forward-Process Aligned Diffusion prediction (\textbf{DiFA}), a training-free framework that reframes inference-time data prediction refinement as a sequential state estimation problem. Rather than reusing past outputs solely for numerical integration, DiFA treats iterative data predictions along the reverse trajectory as correlated observations to build a forward-aligned temporal consensus. Inspired by Kalman filtering, this consensus aggregates historical predictions according to structural consistency and noise-level compatibility. To counteract the over-smoothing tendency of temporal consensus, we introduce a deviation guidance mechanism to adaptively preserve residual details. Empirically, DiFA yields significant improvements on CIFAR-10 and ImageNet across the evaluated metrics, including FID, IS, and FD-DINOv2, demonstrating that aligning inference with the forward statistical structure substantially improves generative fidelity.