Video and image models
These models watch your footage and stills the way an editor would, then write what they find as searchable metadata.
What they find
- Scenes Where a scene starts and ends, what is happening, where, and the mood
- Scene summaries A richer description of the whole scene, not a single moment
- Objects People, vehicles, animals, products and anything else that matters to you
- Logos Brands and sponsors in shot, for exposure and attribution reporting
- On-screen text Titles, captions, signs, scores and kit numbers
- Custom queries Ask your own question of the footage, for example "find shots with a red car"
Less noise, better results
Unique Frame ID analyses the frames that matter instead of every second of footage. That cuts metadata noise by 87%, so your search results are relevant and your processing costs stay low.
Example output
Illustrative example. Real results depend on your content.
| Time | Type | Result |
|---|---|---|
| 00:00:02 | Scene | Coastal road at dusk |
| 00:00:02 | Object | Car, road sign |
| 00:00:07 | On-screen text | "Meridian" |
FAQ
What file types does it support?
A wide range of video and image formats. We will confirm yours in a demo.
Does it recognise faces?
Face recognition is a separate model with extra legal controls.
Where do the results go?
Back into your asset management system, ready to search.