TC-Padé: Trajectory-Consistent Padé Approximation for Diffusion Acceleration

2026-03-03

AI Summary

This work addresses the degradation in generation quality under low-step (20–30 steps) diffusion sampling, which stems from accumulated feature prediction errors and trajectory drift, making it challenging to balance efficiency and fidelity. To this end, we propose a trajectory-consistent feature prediction framework based on Padé rational function approximation, replacing conventional Taylor expansions. Our method integrates historical residual-driven adaptive coefficient modulation with stage-aware (early/mid/late) dynamic perception to accurately model both asymptotic and transitional behaviors during denoising. This approach substantially enhances trajectory consistency and stability at low step counts, achieving up to 2.88× acceleration on DiT-XL/2, FLUX.1-dev, and Wan2.1 while outperforming existing feature caching techniques across multiple metrics, including FID, CLIP score, Aesthetic score, and VBench-2.0.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Feature Construction/Reformulation
• Natural Language Processing: Generation

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Despite achieving state-of-the-art generation quality, diffusion models are hindered by the substantial computational burden of their iterative sampling process. While feature caching techniques achieve effective acceleration at higher step counts (e.g., 50 steps), they exhibit critical limitations in the practical low-step regime of 20-30 steps. As the interval between steps increases, polynomial-based extrapolators like TaylorSeer suffer from error accumulation and trajectory drift. Meanwhile, conventional caching strategies often overlook the distinct dynamical properties of different denoising phases. To address these challenges, we propose Trajectory-Consistent Padé approximation, a feature prediction framework grounded in Padé approximation. By modeling feature evolution through rational functions, our approach captures asymptotic and transitional behaviors more accurately than Taylor-based methods. To enable stable and trajectory-consistent sampling under reduced step counts, TC-Padé incorporates (1) adaptive coefficient modulation that leverages historical cached residuals to detect subtle trajectory transitions, and (2) step-aware prediction strategies tailored to the distinct dynamics of early, mid, and late sampling stages. Extensive experiments on DiT-XL/2, FLUX.1-dev, and Wan2.1 across both image and video generation demonstrate the effectiveness of TC-Padé. For instance, TC-Padé achieves 2.88x acceleration on FLUX.1-dev and 1.72x on Wan2.1 while maintaining high quality across FID, CLIP, Aesthetic, and VBench-2.0 metrics, substantially outperforming existing feature caching methods.

Problem

• diffusion acceleration
• trajectory drift
• feature caching
• low-step sampling
• denoising dynamics

Innovation

• Pade approximation
• diffusion acceleration
• trajectory consistency
• feature caching
• step-aware prediction