Video training data.

Licensed footage for text-to-video, video understanding, and embodied AI, delivered with captions, scene boundaries, and action labels. Sourced from studios, creators, and contributors, or captured to your spec.

What we can supply

Long-form video with dense, time-aligned captions
Instructional and how-to video
Egocentric and wearable-camera footage
Human action and activity video
Audio-visual speech and talking-face footage
Lecture, educational, and presentation video
Product, commerce, livestream, and broadcast content
Specialist footage, such as surgical procedures and sign language

Labels and enrichment

Dense captions
Shot and scene boundaries
Action and step segments
Object and hand tracking
Camera motion
Face tracks and lip sync
Transcript alignment
Quality and synthetic-content scores

Formats and delivery

MP4 or source format with JSON or JSONL sidecars per clip; WebDataset on request. Delivered to AWS S3, Google Cloud Storage, Azure, or your own storage, with a datasheet and provenance records.

Common uses

Text-to-video generation
Video understanding
Temporal grounding
Embodied AI
Avatars and dubbing
Action recognition

Questions

Do you deliver clips or full-length video?

Both. We deliver full-length source or segment it into clips of the length your training needs, with metadata for every clip.

How do you handle footage of identifiable people?

We document the rights basis for footage with identifiable people, and can filter or exclude content to meet biometric-privacy requirements in specific jurisdictions.

Can we set resolution and content requirements?

Yes. Specify minimums for resolution, frame rate, and duration, plus the content mix you want, and we filter or collect to match.

Own data like this? License it through ModelCore