SOURCE-LINKED INTELLIGENCE

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

arXiv · AI, language, vision and robotics · article · Aug 31, 2026 · UTC

Diffusion models have recently advanced text-to-video (T2V) generation, yet they still struggle with fine-grained compositional alignment, such as attribute binding, spatial relations, and object interactions. While reward-based fine-tuning improves alignment, it is susceptible to reward hacking and adapts poorly to new prompt distributions. In this work, we propose NoisEasier, a test-time scaling framework that improves T2V generation through differentiable reward-guided noise optimization without modifying the underlying model. By combining efficient short-step generators with a multi-object

Read original source ↗ Open in workspace

recordType

paper

region

Global