Forecast then Calibrate: Feature Caching as ODE for Efficient Diffusion Transformers

TL;DR

This paper introduces FoCa, a novel ODE-based approach for feature caching in diffusion transformers, significantly improving inference speed while maintaining high generation quality across various tasks.

Contribution

It models feature caching as a feature-ODE solving problem, enabling robust acceleration of diffusion transformers without additional training.

Findings

Achieves near-lossless speedups of 5.50x on FLUX and 6.45x on HunyuanVideo.

Maintains high quality with a 4.53x speedup on DiT.

Demonstrates effectiveness across image synthesis, video generation, and super-resolution tasks.

Abstract

Diffusion Transformers (DiTs) have demonstrated exceptional performance in high-fidelity image and video generation. To reduce their substantial computational costs, feature caching techniques have been proposed to accelerate inference by reusing hidden representations from previous timesteps. However, current methods often struggle to maintain generation quality at high acceleration ratios, where prediction errors increase sharply due to the inherent instability of long-step forecasting. In this work, we adopt an ordinary differential equation (ODE) perspective on the hidden-feature sequence, modeling layer representations along the trajectory as a feature-ODE. We attribute the degradation of existing caching strategies to their inability to robustly integrate historical features under large skipping intervals. To address this, we propose FoCa (Forecast-then-Calibrate), which treats feature caching as a feature-ODE solving problem. Extensive experiments on image synthesis, video generation, and super-resolution tasks demonstrate the effectiveness of FoCa, especially under aggressive acceleration. Without additional training, FoCa achieves near-lossless speedups of 5.50 times on FLUX, 6.45 times on HunyuanVideo, 3.17 times on Inf-DiT, and maintains high quality with a 4.53 times speedup on DiT.