AI Paper Atlas

Research map / Video / Video generation

DisCa: Accelerating Video Diffusion Transformers with Distillation-Compatible Learnable Feature Caching

Chang Zou, Changlin Li, Songtao Liu, Zhao Zhong, Kailin Huang, Linfeng Zhang

CVPR 2026 · Video › Video generation · approach: Efficient video diffusion models · 4 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
restrict the MeanFlow average-velocity interval to [0, R] with R=0.2 (Restricted MeanFlow)MeanFlow distillation of HunyuanVideo, 20 steps · HunyuanVideo VBenchSemantic score: 66.6 -> 70.4Improved
restrict the MeanFlow average-velocity interval to [0, R] with R=0.2 (Restricted MeanFlow)MeanFlow distillation of HunyuanVideo, 10 steps · HunyuanVideo VBenchSemantic score: 60.9 -> 68.2Improved
use R=0.4 instead of R=0.2 in Restricted MeanFlow (ablation)Restricted MeanFlow, 10 steps · HunyuanVideo VBenchSemantic score: 68.2 -> 67.6No clear effect
use Restricted MeanFlow instead of original MeanFlow as the base for the cache predictor (ablation)DisCa · HunyuanVideo VBenchSemantic score: 65.2 -> 69.3Improved
add learnable neural predictor instead of training-free caching (ablation)DisCa on distilled model · HunyuanVideo VBenchSemantic score: 67.3 -> 69.3Improved
add generative adversarial training to the predictor (ablation)DisCa predictor · HunyuanVideo VBenchSemantic score: 68.5 -> 69.3Improved
apply DisCa (learnable caching, N=2) on top of Restricted MeanFlow 20 stepsRestricted MeanFlow 20 steps · HunyuanVideo VBenchSemantic score and latency: 70.4 -> 70.8 semantic; 4.97x -> 7.56x speedImproved
apply training-free caching (TaylorSeer, FORA, TeaCache, PAB, Delta-DiT) to the 10-step distilled modelCFG-distilled HunyuanVideo, 10 steps · HunyuanVideo VBenchSemantic score: 59.0 -> 57.5 (FORA N=6)Hurt
use TaylorSeer forecasting caching on distilled model at high accelerationCFG-distilled HunyuanVideo, 10 steps · HunyuanVideo VBenchSemantic score: 59.0 -> 63.7Improved
use DisCa N=4 at 11.8x speed versus 50-step CFG-distilled modelDisCa · HunyuanVideo VBenchTotal score: 77.9 -> 78.8Improved
reduce sampling steps of CFG-distilled model from 50 to 10 without other changesHunyuanVideo · HunyuanVideo VBenchSemantic score: 66.7 -> 59.0Hurt
use single-tensor learned cache instead of per-layer cachesDisCa · HunyuanVideo peak VRAMPeak VRAM: 130.7GB -> 97.64GBImproved

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