AI Paper Atlas

Research map / Video / Video generation

VMonarch: Efficient Video Diffusion Transformers with Structured Attention

Cheng Liang, Haoxian Chen, Liang Hou, Qi Fan, Gangshan Wu, Xin Tao, Limin Wang

CVPR 2026 · Video › Video generation · approach: Efficient video diffusion models · 3 citations (OpenAlex)

Read the paper

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 first-frame full-attention recomputation (ablation)VMonarch · VBench training-freePSNR vs full attention: 11.65 -> 12.59Improved
add first-frame recomputation (fine-tuned) (ablation)VMonarch · VBench fine-tunedvalidation loss: 1.27 -> 1.25Improved
replace recomputation with attention-sink averaging (ablation)VMonarch · VBench fine-tunedvalidation loss: 1.25 -> 1.78Hurt
increase Monarch iterations from 2 to 7 (fine-tuned) (ablation)VMonarch · VBench fine-tunedDynamic Degree: 68.06 -> 36.11Hurt
increase Monarch iterations from 1 to 3 (training-free) (ablation)VMonarch · VBench training-freePSNR vs full attention: 11.18 -> 12.21Improved
use block size of two frames instead of one (ablation)VMonarch · VBench fine-tunedvalidation loss: 1.25 -> 1.37Hurt
fine-tune 1500 steps (training-free to tuned) (ablation)VMonarch · VBench Dynamic DegreeDynamic Degree: 54.17% -> 62.50%Improved
VMonarch vs VSA at 90% sparsity, training-freeVMonarch · VBench training-freeAesthetic Quality: 42.86% -> 63.65%Improved
VMonarch vs VSA on end-to-end time (61x448x832)VMonarch · Inference timetime (s): 49.9 -> 47.7Improved
VMonarch vs full attention on long video time (141 frames)VMonarch · Inference timetime (s): 214.2 -> 102.7Improved
VMonarch vs full attention on Dynamic Degree after tuningVMonarch · VBench 61x448x832Dynamic Degree: 68.06% -> 62.50%Hurt
VMonarch vs VSA on Wan2.1-14B qualityVMonarch · VBench 93x704x1280Aesthetic Quality: 66.40% -> 65.91%Mixed

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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