Awesome World Model Hub Papers · Datasets · Projects
← Back to datasets

World-R1

2026 Datasets
Preview for World-R1

Dataset Analysis

World-R1 uses reinforcement learning to align text-to-video generation with 3D constraints, improving geometric consistency without architectural changes.

Provenance

Collected from papers.

Derived from paper: World-R1: Reinforcing 3D Constraints for Text-to-Video Generation

Resources

Tags

text-to-video generation3D constraintsreinforcement learningFlow-GRPOgeometric consistencyworld simulationCV