Search papers, labs, and topics across Lattice.
This paper introduces ResilPhase, a novel framework for accelerating diffusion models by reformulating inference as stable macro-trajectory extrapolation in ODE space. By aligning forecasting with the model's Global Drift and employing a derivative-free barycentric Lagrange extrapolator, the method effectively mitigates issues related to feature inconsistency and derivative instability. Experimental results on FLUX.1-dev and HunyuanVideo show that ResilPhase achieves state-of-the-art fidelity even under high acceleration ratios, addressing the critical challenge of inference latency in diffusion models.
ResilPhase achieves high-fidelity diffusion model acceleration by eliminating derivative instability, setting a new standard for inference speed without sacrificing quality.
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.
