CalibCache: Cache-to-Full Feature Dynamics Calibration for Diffusion Transformers CalibCache, a lightweight offline calibration method for diffusion transformers, cuts FLUX.1-dev LPIPS from 0.563 to 0.429 at cache interval N=5 while delivering a 2.86x denoising speedup over full-compute sampling, according to the arXiv paper 2611.25296v1. The method fits scalar acceleration maps offline by centered least squares from a single pair of full-compute and naive-cache trajectories, then applies them during sampling without extra full-network evaluations or online optimization. Across FLUX.1-dev and Qwen-Image on DrawBench and DiT-XL/2 on ImageNet-256, CalibCache improved fidelity or generation quality over TaylorSeer under 20-step sampling at cache intervals N=3, 4 and 5, and lowered DiT-XL/2 FID from 16.39 to 10.37 relative to TaylorSeer. arXiv:2611.25296v1 Announce Type: new Abstract: Diffusion Transformers generate high-quality images, but iterative sampling remains computationally expensive. Feature caching reduces the number of full network evaluations, yet its approximation error can become more severe under short sampling schedules, where fewer full-compute steps and a coarser sampling grid can make history-based extrapolation less reliable. In this paper, we propose CalibCache, a lightweight offline calibration method that corrects systematic discrepancies between cached and full-compute feature dynamics. Specifically, CalibCache calibrates feature acceleration, defined as the second-order finite difference across sampling steps. Using only a single pair of full-compute and naive-cache trajectories at each setting, CalibCache fits scalar acceleration maps offline by centered least squares. These maps are then applied during sampling without additional full-network evaluations or online optimization. Across FLUX.1-dev and Qwen-Image on DrawBench and DiT-XL/2 on ImageNet-256, CalibCache consistently improves fidelity or generation quality over TaylorSeer under 20-step sampling at cache intervals $N=3,4,5$. At $N=5$, it reduces FLUX.1-dev LPIPS from $0.563$ to $0.429$ while achieving a $2.86\times$ denoising speedup over full-compute sampling, and lowers DiT-XL/2 FID from $16.39$ to $10.37$ relative to TaylorSeer.