AI Paper Atlas

Research map / Video / Video generation

MoGA: Mixture-of-Groups Attention for End-to-End Long Video Generation

Weinan Jia, Yuning Lu, Mengqi Huang, Hualiang Wang, Binyuan Huang, Nan Chen, Mu Liu, Jidong Jiang, Zhendong Mao

ICLR 2026 · Video › Video generation · approach: Efficient video diffusion models · 21 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
replace block-sparse routing (VMoBA) with MoGA token routerWan2.1-14B · 5-second single-shotsubject consistency: 0.8605 -> 0.9699Improved
apply MoGA at 71.25% sparsity vs full attentionWan2.1-14B · 5-second single-shotsubject consistency: 0.9611 -> 0.9699Improved
apply MoGA vs full attentionWan2.1-14B · 5-second single-shotbackground consistency: 0.9560 -> 0.9542No clear effect
use MoGA instead of EchoShot full attentionWan2.1-1.3B · 10-second multi-shotcross-shot DINO: 0.5961 -> 0.6623Improved
use MoGA instead of EchoShot full attentionWan2.1-1.3B · 10-second multi-shotmotion smoothness: 0.9939 -> 0.9919Hurt
end-to-end MoGA vs multistage IC-LoRA+WanWan2.1-14B · 30-second multi-shotsubject consistency: 0.8946 -> 0.9572Improved
increase routing groups from 1 to 4 (ablation)MoGA Wan2.1-1.3B · 10-second multi-shotcross-shot DINO: 0.8206 -> 0.8672Improved
increase routing groups from 4 to 16 (ablation)MoGA Wan2.1-1.3B · 10-second multi-shotcross-shot DINO: 0.8672 -> 0.8569Hurt
increase routing groups from 1 to 16 (ablation)MoGA Wan2.1-1.3B · 10-second multi-shotcross-shot CLIP: 0.5919 -> 0.6896Improved
use MoGA with M=5 instead of full attentionWan2.1-1.3B · 30-second videoattention PFLOPs: 6.94 -> 2.26Improved
use MoGA alone without STGA (ablation)MoGA · long video generationvisual quality (qualitative)Hurt
use STGA alone without MoGA (ablation)MoGA · long video generationcross-shot consistency (qualitative)Hurt

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 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