AI Paper Atlas

Research map / Video / Video generation

NoisEasier: Test-Time Noise Optimization for Text-to-Video Generation

Yujiang Pu, Yu Kong

ECCV 2026 · Video › Video generation · approach: Compositional text-to-video diffusion · 1 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 NoisEasier test-time noise optimizationT2V-Turbo (MS) · VBenchMultiple Objects: 54.50 -> 76.62Improved
add NoisEasierT2V-Turbo (VC2) · VBenchSpatial Relation: 39.95 -> 52.61Improved
add NoisEasierT2V-Turbo (MS) · VBenchMotion Smoothness: 95.51 -> 95.29Hurt
add NoisEasierAnimateLCM · T2V-CompBenchConsistent Attribute Binding: 0.6746 -> 0.7807Improved
add NoisEasierT2V-Turbo (VC2) · T2V-CompBenchDynamic Attribute: 0.0113 -> 0.0097Hurt
optimize full trajectory (z_T and perturbations) instead of initial noise only (ablation)NoisEasier · VBench Overall Consistencyscore: T2V-Turbo MS 29.67 -> 32.05Improved
optimize full trajectory instead of initial only (ablation)NoisEasier · VBench Dynamic Degreescore: AnimateLCM 30.83 -> 30.56; MS 66.94 -> 71.39Mixed
add ImageReward only vs combined rewards (ablation)NoisEasier · VBench Dynamic Degreescore: 70.83 -> 61.11Hurt
combine all rewards instead of single reward (ablation)NoisEasier · VBench Overall Consistencyscore: 36.21 (ViCLIP only) vs 32.05 combined; Dynamic 63.89 -> 71.39Improved
add negative-aware reward calibration (ablation)NoisEasier · VBench Colorscore: 93.29 -> 94.38Improved
use random POS-perturbation negatives instead of LLM negatives (ablation)NoisEasier · VBench Spatial Relationscore: 47.95 -> 46.39 (vs no negs)Hurt
use noise optimization instead of best-of-n samplingNoisEasier · VBench Multiple Objectsscore: 80.06 (BoS) vs 76.62Hurt

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