ProCache: Constraint-Aware Feature Caching with Selective Computation for Diffusion Transformer Acceleration

TL;DR

ProCache is a dynamic, constraint-aware feature caching framework for diffusion transformers that improves acceleration efficiency by aligning caching with temporal dynamics and selectively recomputing critical features, with minimal quality loss.

Contribution

ProCache introduces a novel non-uniform caching pattern and selective computation strategy tailored to diffusion transformer dynamics, enabling effective acceleration without retraining.

Findings

Achieves up to 2.90x acceleration on diffusion models.

Maintains high generation quality with negligible degradation.

Outperforms prior caching methods significantly.

Abstract

Diffusion Transformers (DiTs) have achieved state-of-the-art performance in generative modeling, yet their high computational cost hinders real-time deployment. While feature caching offers a promising training-free acceleration solution by exploiting temporal redundancy, existing methods suffer from two key limitations: (1) uniform caching intervals fail to align with the non-uniform temporal dynamics of DiT, and (2) naive feature reuse with excessively large caching intervals can lead to severe error accumulation. In this work, we analyze the evolution of DiT features during denoising and reveal that both feature changes and error propagation are highly time- and depth-varying. Motivated by this, we propose ProCache, a training-free dynamic feature caching framework that addresses these issues via two core components: (i) a constraint-aware caching pattern search module that generates non-uniform activation schedules through offline constrained sampling, tailored to the model's temporal characteristics; and (ii) a selective computation module that selectively computes within deep blocks and high-importance tokens for cached segments to mitigate error accumulation with minimal overhead. Extensive experiments on PixArt-alpha and DiT demonstrate that ProCache achieves up to 1.96x and 2.90x acceleration with negligible quality degradation, significantly outperforming prior caching-based methods.