Built for frontier AI teams
Built for frontier AI teams
Video datasets
for frontier models.
Built for teams training large video and world models across robotics, ads, and avatars.
High fidelity data
Built for training
Easy integration
FROM THE LAB
FROM THE LAB
FROM THE LAB
We collect, clean, and evaluate video so your team can focus on models.
Curated video data
Curated video data
Curated video data
Benchmarks and metrics
Benchmarks and metrics
Benchmarks and metrics
Production ready formats
Production ready formats
Production ready formats
READY TO LICENSE
READY TO LICENSE
READY TO LICENSE
Datasets, built for world models, robotics, and generative video systems
WorldScenes 100K
Open world short clips across cities, homes, transit, nature, and retail, three to ten seconds, 24 to 30 fps, multilingual captions, entities, activities, depth and optical flow on subsets, panoptic segmentation on stratified splits.
PersonaFrames 30K
PhysicsBench 10K
AdCreative 50K
EgoHands 12K
WorldScenes 100K
Open world short clips across cities, homes, transit, nature, and retail, three to ten seconds, 24 to 30 fps, multilingual captions, entities, activities, depth and optical flow on subsets, panoptic segmentation on stratified splits.
PersonaFrames 30K
PhysicsBench 10K
AdCreative 50K
EgoHands 12K
WorldScenes 100K
Open world short clips across cities, homes, transit, nature, and retail, three to ten seconds, 24 to 30 fps, multilingual captions, entities, activities, depth and optical flow on subsets, panoptic segmentation on stratified splits.
PersonaFrames 30K
PhysicsBench 10K
AdCreative 50K
EgoHands 12K
Train models that understand how the world really moves
BUILT FOR FRONTIER TEAMS
BUILT FOR FRONTIER TEAMS
BUILT FOR FRONTIER TEAMS
Let us show your models the world.
Curated video data
High quality clips reviewed, filtered, and organized by our team, not scraped at random.
Curated video data
High quality clips reviewed, filtered, and organized by our team, not scraped at random.
Curated video data
High quality clips reviewed, filtered, and organized by our team, not scraped at random.
Built from real failures
We start from where your models break and design datasets to stress those cases.
Built from real failures
We start from where your models break and design datasets to stress those cases.
Built from real failures
We start from where your models break and design datasets to stress those cases.
Clear rights and safety
Rights, consent, and safety checks handled so legal and security can move quickly.
Clear rights and safety
Rights, consent, and safety checks handled so legal and security can move quickly.
Clear rights and safety
Rights, consent, and safety checks handled so legal and security can move quickly.
Training ready formats
Seamlessly customize layouts and features to match any style.
Training ready formats
Seamlessly customize layouts and features to match any style.
Training ready formats
Seamlessly customize layouts and features to match any style.
Continuously refreshed
New domains, edge cases, and updated evaluators as your products, models, and data needs change.
Continuously refreshed
New domains, edge cases, and updated evaluators as your products, models, and data needs change.
Continuously refreshed
New domains, edge cases, and updated evaluators as your products, models, and data needs change.