AI Paper Atlas

Research map / Video / Video generation

Adaptive Video Distillation: Mitigating Oversaturation and Temporal Collapse in Few-Step Generation

Yuyang You, Yongzhi Li, Jiahui Li, Yadong Mu, Quan Chen, Peng Jiang

CVPR 2026 · Video › Video generation · approach: Efficient video diffusion models · 5 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
add temporal regularization to DMD (ablation)DMD 1.3B · VBench Dynamic DegreeDynamic Degree: 72.22 -> 100.00Improved
add temporal regularization to DMD (ablation)DMD 1.3B · VBench Instance PreservationInstance Preservation: 88.88 -> 85.38Hurt
add naive regression loss on top of temporal regularization (ablation)DMD+TR 1.3B · VBench Instance PreservationInstance Preservation: 85.38 -> 83.04Hurt
add naive regression loss on top of temporal regularization (ablation)DMD+TR 1.3B · VBench Dynamic DegreeDynamic Degree: 100.00 -> 78.61Hurt
replace naive regression loss with adaptive regression loss (ablation)DMD+TR+RegLoss 1.3B · VBench Instance PreservationInstance Preservation: 83.04 -> 92.39Improved
replace naive regression loss with adaptive regression loss (ablation)DMD+TR+RegLoss 1.3B · VBench Dynamic DegreeDynamic Degree: 78.61 -> 99.72Improved
add inference-time frame interpolation (ablation)DMD+TR+AdaLoss 1.3B · Inference timetime (s): 10.8s -> 7.8sImproved
add inference-time frame interpolation (ablation)DMD+TR+AdaLoss 1.3B · VBench Instance PreservationInstance Preservation: 92.39 -> 91.81No clear effect
add inference-time frame interpolation (ablation)DMD+TR+AdaLoss 1.3B · VBench Dynamic DegreeDynamic Degree: 99.72 -> 97.77No clear effect
use full method instead of DMD baselineWan2.1-T2V-1.3B 4-step · VBench2 Totalscore: 53.63 -> 55.08Improved
use full method instead of DMD baselineWan2.1-T2V-14B 4-step · VBench2 Totalscore: 56.87 -> 59.06Improved
use full method instead of DMD baselineWan2.1-T2V-14B 4-step · VBench1 Totalscore: 79.63 -> 82.57Improved

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