IdealCache speeds up diffusion transformers with better cache schedules
IdealCache: Rethinking Cache Scheduling in Diffusion Transformers via Ideal Trajectories
Artificial Intelligence
Summary
Diffusion transformers create images or videos step by step, and usually save some calculations to speed up this process. But usual methods try to copy a fixed step-by-step path, which limits quality and speed. The authors propose IdealCache, which instead finds the best caching strategy by looking at many possible paths to the final result, not just one. This approach significantly speeds up generation without losing quality and even improves results in some editing tasks.
What this means in practice
- •For ai application developers: Accelerate image and video generation models by using IdealCache’s optimized caching without retraining, improving efficiency and output quality.
- •For digital content editors: Enhance semantic consistency in image editing tasks by applying IdealCache’s schedule to maintain higher quality during generation.
Authors
Shang Xue, Li Zhu, Ping Chen
Abstract
Feature caching accelerates Diffusion Transformers by reusing intermediate computations across denoising timesteps, yet existing methods share a fundamental blind spot: they treat the standard reference denoising trajectory as the optimisation target, approximating its per-step states and thus inheriting its output as an implicit quality ceiling. We argue that this target is arbitrary -- what truly matters is the ideal result, not the intermediate path taken to reach it. We propose IdealCache, a training-free framework that abandons step-wise imitation of the reference trajectory. Instead, it anchors caching on a per-input ensemble of ideal trajectories constructed from candidate target results, aggregates their geometry into a single graph, and discovers optimal non-uniform schedules via a constrained shortest path on this graph. By decoupling from the reference process, IdealCache breaks the performance bottleneck of trajectory-approximation; our schedules can attain near-lossless fidelity in generation and even surpass on task-specific metrics (e.g., maintaining higher semantic consistency in image editing). Without retraining, IdealCache achieves 6.35x speedup on Qwen-Image, 5.84x on Qwen-Image-Edit, and 4.32x on HunyuanVideo, with competitive or superior quality across tasks.