ResilPhase: Plug-and-Play Phase Mapping and Noise-Resilient Macro-Trajectory Extrapolation for Diffusion Acceleration

2026-06-25

AI Summary

This work addresses the severe degradation in generation quality of diffusion models under high acceleration ratios, which stems from spatial error accumulation due to feature inconsistency and numerical instability, amplified derivative noise, and higher-order instabilities. To mitigate these issues, the authors propose a macroscopic trajectory extrapolation framework operating in the ordinary differential equation (ODE) space. This approach avoids intermediate feature prediction by aligning global drift and enhances robustness through derivative-free barycentric Lagrange extrapolation combined with a bounded phase mapping. Evaluated on FLUX.1-dev and HunyuanVideo, the method significantly outperforms existing cache-and-predict acceleration techniques, achieving state-of-the-art generation quality and stability at high acceleration ratios.

Abstract

The adoption of powerful diffusion models is hindered by their significant inference latency. Recent ``cache-then-forecast'' schemes alleviate this issue by accelerating DiTs using derivative-based polynomials, but they suffer from severe quality degradation at high acceleration ratios. Our analysis reveals its root cause: the discrete extrapolation performed on representations that are misaligned with the continuous diffusion trajectory and are numerically unstable. Thus, accelerated DiTs suffer from accumulated spatial errors, noisy derivative amplification, and high-order instability. We therefore reformulate accelerated inference as stable macro-trajectory extrapolation in ordinary differential equation (ODE) space. Instead of predicting intermediate features, we align forecasting with the model's Global Drift (GD), i.e., the end-to-end state evolution, thereby eliminating feature inconsistency and memory overhead. However, even this smooth macro-trajectory remains vulnerable to the derivative fallacy: its higher-order temporal derivatives are intrinsically noisy. Thus, we introduce a derivative-free barycentric Lagrange extrapolator to effectively bypass derivative instability and approximation error. We further propose a bounded Phase Mapping that regularizes the extrapolation domain, suppressing oscillatory error growth. These elements collectively constitute ResilPhase, a noise-resilient acceleration framework. Experiments on FLUX.1-dev and HunyuanVideo demonstrate state-of-the-art fidelity under aggressive acceleration ratios.

Problem

• diffusion acceleration
• inference latency
• trajectory extrapolation
• numerical instability
• noise resilience

Innovation

• macro-trajectory extrapolation
• derivative-free extrapolation
• phase mapping
• noise-resilient acceleration
• diffusion model inference