WorldScore
by Stanford University (Haoyi Duan, Hong-Xing Yu, Sirui Chen, Li Fei-Fei, Jiajun Wu)
What is WorldScore? WorldScore is an image & vision eval made by Stanford University (Haoyi Duan, Hong-Xing Yu, Sirui Chen, Li Fei-Fei, Jiajun Wu). A unified benchmark for world generation across 3D scene models, 4D generation and video models. World generation is broken into next-scene tasks with explicit camera-trajectory layouts. It has 3,000 test examples and scores controllability (camera, object, content), quality (3D / photometric / style consistency and subjective quality) and dynamics (motion accuracy, magnitude, smoothness). ICCV 2025 (arXiv:2504.00983). Leaderboard on Hugging Face (Howieeeee/WorldScore_Leaderboard).
Editor note
A unified benchmark for world generation across 3D scene models, 4D generation and video models. World generation is broken into next-scene tasks with explicit camera-trajectory layouts. It has 3,000 test examples and scores controllability (camera, object, content), quality (3D / photometric / style consistency and subjective quality) and dynamics (motion accuracy, magnitude, smoothness). ICCV 2025 (arXiv:2504.00983). Leaderboard on Hugging Face (Howieeeee/WorldScore_Leaderboard).
- What it tests
- Camera-controllable world generation: controllability, quality and dynamics across 3D, 4D and video models.
- Original source
- haoyi-duan.github.io/WorldScore/
- Last updated
- Repo last push 2026-07-23 (GitHub API, earlier read 14 Nov 2026); LICENSE re-read 14 Nov 2026
- License
- MIT (GitHub LICENSE, re-read 14 Nov 2026)
Related image & vision evals
Discussion
No one has discussed WorldScore yet. Have you run it? Share what it caught, what it missed or how you set it up.