English

InvarDiff: Cross-Scale Invariance Caching for Accelerated Diffusion Models

Computer Vision and Pattern Recognition 2025-12-08 v1 Distributed, Parallel, and Cluster Computing Machine Learning

Abstract

Diffusion models deliver high-fidelity synthesis but remain slow due to iterative sampling. We empirically observe there exists feature invariance in deterministic sampling, and present InvarDiff, a training-free acceleration method that exploits the relative temporal invariance across timestep-scale and layer-scale. From a few deterministic runs, we compute a per-timestep, per-layer, per-module binary cache plan matrix and use a re-sampling correction to avoid drift when consecutive caches occur. Using quantile-based change metrics, this matrix specifies which module at which step is reused rather than recomputed. The same invariance criterion is applied at the step scale to enable cross-timestep caching, deciding whether an entire step can reuse cached results. During inference, InvarDiff performs step-first and layer-wise caching guided by this matrix. When applied to DiT and FLUX, our approach reduces redundant compute while preserving fidelity. Experiments show that InvarDiff achieves 22-3×3\times end-to-end speed-ups with minimal impact on standard quality metrics. Qualitatively, we observe almost no degradation in visual quality compared with full computations.

Keywords

Cite

@article{arxiv.2512.05134,
  title  = {InvarDiff: Cross-Scale Invariance Caching for Accelerated Diffusion Models},
  author = {Zihao Wu},
  journal= {arXiv preprint arXiv:2512.05134},
  year   = {2025}
}

Comments

8 pages main, 8 pages appendix, 16 figures, 5 tables. Code: https://github.com/zihaowu25/InvarDiff

R2 v1 2026-07-01T08:10:08.992Z