LinCa: Accelerating Diffusion Models via Learnable Decomposed Feature Caching

Dix001Busion models have achieved remarkable success in im-age and video generation, yet the high computational cost of iterativesampling remains a critical bottleneck for practical deployment. Featurecaching has emerged as a promising acceleration paradigm by reusingor predicting intermediate features across timesteps. However, existingtraining-free methods apply uniform prediction strategies that cannotadapt to the heterogeneous feature dynamics, causing signix001Ccant qualitydegradation under high acceleration ratios. We propose LinCa , a fea-ture caching framework based on learnable invertible networks. LinCadecomposes cached features into sub-components with distinct conti-nuity properties via a lightweight invertible network and applies dif-ferentiated prediction orders matched to each component. The strictinvertibility guarantees lossless reconstruction back to the original fea-ture space, forming a unix001Ced Decompose-Predict-Reconstruct pipeline.By training separate predictors for dix001Berent models and timestep seg-ments, LinCa adapts to heterogeneous feature dynamics. Experiments onFLUX, Qwen-Image, and HunyuanVideo demonstrate that LinCa, withless than 0.2% additional parameters, signix001Ccantly outperforms exist-ing methods and maintains near-lossless quality at 5-7× speedup. Code:https://github.com/QHR69/LinCa.