BeziCast speeds up diffusion image and video generation by forecasting
Tikhonov-Stabilized Bezier Representation Forecasting for Training-free Diffusion Acceleration
Computer Vision and Pattern Recognition
Summary
Generating images and videos with diffusion models is usually slow because the computer needs to process many steps repeatedly. The paper presents BeziCast, a method that speeds this up without retraining the model by predicting intermediate results using smooth curves called Bézier trajectories. This approach avoids instability seen in previous prediction methods and reduces error buildup by focusing on outputs close to the final step. BeziCast demonstrates faster generation with better quality preservation than earlier acceleration techniques.
What this means in practice
- •For image generation engineers: Speed up image generation pipelines by forecasting denoising outputs without retraining or changing existing diffusion models.
- •For video production teams: Accelerate video generation workflows using diffusion models by predicting intermediate states for faster rendering while maintaining quality.
Authors
Lei Zhu, Mujie Lin, Ruochong Zheng, Guangyi Wang, Hao Li, Peng Jin, Chang Liu, Jie Chen
Abstract
Diffusion models, particularly Diffusion Transformers, achieve strong image and video generation quality but remain expensive at inference time due to repeated denoiser evaluations. Feature caching offers a deployment-friendly acceleration strategy by reusing or predicting intermediate representations without retraining the generator or modifying the sampler. However, existing cache-then-forecast methods often rely on Taylor-style finite-difference extrapolation, which can become unstable over longer cache intervals, and typically predict local module outputs whose errors may accumulate through subsequent denoiser blocks. We propose \textbf{BeziCast}, a training-free diffusion acceleration framework that forecasts output-proximal denoising representations using low-order B'ezier trajectories. BeziCast estimates B'ezier control points via Tikhonov-stabilized fitting, providing a smooth, capacity-controlled temporal parameterization that avoids explicit high-order derivative estimation and decouples trajectory capacity from cache-query density. To moderate aggressive extrapolation, we further derive equivalent forecasting weights and introduce a convex-hull-inspired guardrail that detects high-risk predictions and softly pulls them toward a conservative simplex-projected estimate. Extensive evaluations across advanced image and video diffusion models demonstrate the effectiveness of BeziCast. In particular, BeziCast accelerates FLUX.1 by up to (4.79\times) and HunyuanVideo by up to (4.11\times), while preserving substantially better generation quality than competing acceleration baselines.