ThinkVid: Benchmarking Visual Reasoning in Video Generative Models

Harold Haodong Chen

Co-first author

Wenjie Shu

Co-first author

,

Disen Lan

Co-first author

,

Qingyang Liu

,

Zihan Wang

,

And 12 Other Authors

· 1 min read

Abstract

A benchmark for evaluating visual reasoning ability of video generative models.

Type

Conference paper

Publication

CVPR 2026

Accepted at CVPR 2026.

Last updated on Jan 1, 2026

Peer-Reviewed Video Generation Evaluation Reasoning

Wenjie Shu

Authors

Wenjie Shu (he/him)

Researcher

← Show, Don't Tell: Morphing Latent Reasoning into Image Generation Jan 1, 2026

Enhancing Diffusion-based Restoration Models via Difficulty-Adaptive Reinforcement Learning with IQA Reward Nov 1, 2025 →