AI Paper Atlas

Research map / Video / Video generation

Model Reveals What to Cache: Profiling-Based Feature Reuse for Video Diffusion Models

Xuran Ma, Yexin Liu, Yaofu Liu, Xianfeng Wu, Mingzhe Zheng, Zihao Wang, Ser-Nam Lim, Harry Yang

ICCV 2025 · Video › Video generation · approach: Efficient video diffusion models · 11 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
reuse background-focused blocks instead of foreground-focused blocks (ablation)ProfilingDiT · HunyuanVideo 720PLPIPS: 0.1449 -> 0.1203Improved
reuse background-focused blocks instead of foreground-focused blocks (ablation)ProfilingDiT · HunyuanVideo 720PPSNR: 24.23 -> 26.44Improved
reuse background-focused blocks instead of foreground-focused blocks (ablation)ProfilingDiT · HunyuanVideo 720Pspeedup: 1.57x -> 1.87xImproved
reuse background-focused blocks instead of foreground-focused blocks (ablation)ProfilingDiT · HunyuanVideo 720PFID: 58.53 -> 41.10Improved
use background-only reuse instead of Split background+foreground reuse (ablation)ProfilingDiT · HunyuanVideo 720PLPIPS: 0.1753 -> 0.1203Improved
use background-only reuse instead of Alternate background+foreground reuse (ablation)ProfilingDiT · HunyuanVideo 720PLPIPS: 0.1875 -> 0.1203Improved
use decreasing (Stepwise) reuse intervals instead of increasing (Step Inverse) (ablation)ProfilingDiT background reuse · HunyuanVideo 720PLPIPS: 0.2529 -> 0.1203Improved
use decreasing (Stepwise) reuse intervals instead of increasing (Step Inverse) (ablation)ProfilingDiT background reuse · HunyuanVideo 720PPSNR: 24.49 -> 26.44Improved
use decreasing (Stepwise) reuse intervals instead of fixed (Step Average) (ablation)ProfilingDiT background reuse · HunyuanVideo 720PLPIPS: 0.1571 -> 0.1203Improved
use decreasing (Stepwise) reuse intervals instead of fixed (Step Average) (ablation)ProfilingDiT foreground reuse · HunyuanVideo 720PLPIPS: 0.1303 -> 0.1449Hurt
increase GPU count from 1 to 8 (ablation)ProfilingDiT · HunyuanVideo 720P 129 framesspeedup: 1.87x -> 1.61xHurt
replace TeaCache with ProfilingDiTHunyuanVideo · HunyuanVideo 720P (vs TeaCache slow)PSNR: 21.91 -> 26.44Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy