Flow matching speed improved by segment advancement with shared lookahead
ReAL: Accelerating Flow Matching through Segment Advancement with Shared Lookahead
Computer Vision and Pattern Recognition
Summary
Generating high-quality images or videos with flow-matching models is slow because the neural network has to be used many times. The authors found a way to guess how far to jump ahead without extra checks, by using information they already have about the direction changes. This lets them skip some calculation steps while keeping the output quality nearly the same. Their method, called ReAL, speeds up generation by about five times without needing extra training.
What this means in practice
- •For image generation engineers: Accelerate sampling in image generation models to reduce compute time while maintaining output quality.
- •For video production tool developers: Implement faster neural video generation pipelines by reducing the number of required model evaluations during sampling.
Authors
Xuanhua Yin, Chuanzhi Xu, Haoxian Zhou, Shunqi Mao, Weidong Cai
Abstract
Flow-matching models generate high-quality images and videos, but repeated neural network evaluations make sampling expensive. Skipping evaluations reduces this cost by extending an available velocity estimate over a longer span. However, local velocity agreement alone does not determine a suitable span, and checking each candidate endpoint adds costly model calls. We introduce ReAL, a training-free sampler that selects how far to advance using one shared lookahead. Our key insight is that the discrepancy between uncorrected and lookahead-corrected endpoint proposals can be computed directly from the observed velocity mismatch and the candidate span beyond the lookahead. This relation provides a span-dependent selection criterion without additional endpoint evaluations. The same lookahead selects the longest passing candidate span, corrects the accepted update, and supplies its velocity as the next starting estimate. After initialization, each regular iteration requires only one fresh evaluation. ReAL uses the pretrained velocity output and original noise schedule, with no additional training or access to internal features. Experiments cover four image-generation backbones, video generation, and image editing. ReAL achieves 4.91x measured speedup on FLUX.1-dev while retaining 97.0% of dense mean ImageReward. On HunyuanVideo, it achieves a 5.49x speedup while maintaining a VBench score close to that of dense sampling.