AI Paper Atlas

Research map / Video / Video generation

VideoFlexTok: Flexible-Length Coarse-to-Fine Video Tokenization

Andrei Atanov, Jesse Allardice, Roman Bachmann, Oğuzhan Fatih Kar, R Devon Hjelm, David Griffiths, Peter Fu, Afshin Dehghan, Amir Zamir

ICML 2026 · Video › Video generation · approach: Compositional text-to-video diffusion · 0 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
use VideoFlexTok with 160 tokens instead of 1024-1280 token grid tokenizersclass-to-video AR model · Kinetics-600gFVD: 87.5 -> 80.0Improved
use VideoFlexTok vs Omnitokenizerclass-to-video AR model · Kinetics-600class alignment score: 0.858 -> 0.833Hurt
use VideoFlexTok vs LARP for reconstructiontokenizer · Kinetics-600rFVD: 42.1 -> 48.7No clear effect
generate fewer VideoFlexTok tokens per frame with the AR modelclass-to-video AR model · Kinetics-600model size / training tokens for comparable performance: 5-10x smaller model or 5-10x fewer training tokensImproved
use VideoFlexTok with adjusted token count vs 3D grid tokenizertext-to-video AR model · Panda70Mcompute for comparable gFVD/ViCLIP: order of magnitude less computeImproved
train on 32 tokens per frame for 10-second videostext-to-video AR model · Panda70M 81-frametoken count per video: 5376 -> 672Improved
replace time-causal 2D registers with flat 1D registers (ablation)VideoFlexTok · Kinetics-600rFVD: 69.9 -> 48.9Improved
replace time-causal 2D registers with flat 1D registers (ablation)VideoFlexTok · Kinetics-600 class-to-videogFVD: 287.6 -> 352.1Hurt
use full self-attention instead of time-causal decoder (ablation)VideoFlexTok (32 tokens) · Kinetics-600rFVD: 80.9 -> 58.3Improved
use full self-attention instead of time-causal decoder (ablation)VideoFlexTok (32 tokens) · Kinetics-600 class-to-videogFVD: 175.1 -> 211.5Hurt

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