AI Paper Atlas

Research map / Video / Video generation

Importance-Based Token Merging for Efficient Image and Video Generation

Haoyu Wu, Jingyi Xu, Hieu Le, Dimitris Samaras

ICCV 2025 · Video › Video generation · approach: Compositional text-to-video diffusion · 16 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
importance-based destination selection instead of random (ToMeSD)Stable Diffusion 2 token merging · MS-COCO 30K, r=0.75FID: 20.89 -> 17.75Improved
importance-based merging at low merging ratioStable Diffusion 2 token merging · MS-COCO 30K, r=0.3FID: 12.20 -> 12.20No clear effect
importance-based merging on diffusion transformerPixArt-alpha · MS-COCO, r=0.7FID: 95.46 -> 50.76Improved
importance-based merging on multi-view diffusionZero123++ · GSO, r=0.75PSNR: 13.12 -> 14.58Improved
importance-based merging on multi-view diffusionZero123++ · GSO, r=0.75SSIM: 0.789 -> 0.784No clear effect
importance-based merging on video diffusionAnimateDiff v3 · VBench, r=0.75VBench total: 78.59 -> 79.12Improved
merge tokens in temporal layers of video modelAnimateDiff v3 · VBenchgenerated dynamicsHurt
choose destination tokens by top-k importance instead of random from pool (ablation)importance-based merging · MS-COCO SD2, r=0.3FID: 12.20 -> 12.56Hurt
draw independent tokens only from the important pool (ablation)importance-based merging · MS-COCO SD2, r=0.7FID: 16.43 -> 16.22Improved
use CFG instead of cross-attention maps as importance signal (ablation)importance-based merging · MS-COCO SD2, r=0.75FID: 18.17 -> 17.75Improved
add important token pool (p from 0 to 0.4) (ablation)importance-based merging · MS-COCO SD2, r=0.3FID: 12.87 -> 12.19Improved
use importance-based instead of random mergingStable Diffusion 2 · inference cost, r=0.7latency and TFLOPs: 5.7 -> 5.8 s latency; 3.36 TFLOPsNo clear effect

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