Know what every video is actually about. Send video URLs and get back IAB categories with IDs and scores, plus 35 key terms (topics, people, places and themes) with their aliases. Your search, recommendations, targeting and brand-safety models can use the output directly.

Why titles and tags aren't enough

Most video libraries are labeled by whoever uploaded them: a title, a genre and a few tags. Those labels rarely describe what is on screen or what is said, and every model trained on them inherits the error. Classification from the video itself gives you labels you can trust.

How it works

  1. Send video URLs. Batches of up to 500 public video or page URLs per request. Page URLs are scanned for the videos on them.
  2. Pick the signals. Classification returns IAB categories with the transcript, plus semantic terms.
  3. Use the JSON. Store it with each video, index it, or feed it to your own models.

What comes back

{"type": "TRANSCRIPTION", "iabCategorization": [{"uniqueId": "337", "version": "2.2", "tier1": "Movies", "tier2": "World Movies", "score": 0.156}]}
{"type": "SEMANTIC_ANALYSIS", "result": {"terms": [{"term": "Hypnosis", "type": "topic", "aliases": ["hypnotism"]}]}}

What teams build with it

  • Categorization for targeting and brand-safety models: train them on what the video is actually about.
  • Content packaging: group a library by real content so advertisers can buy it.
  • Search and discovery: make a video library searchable by topic, person and place.

Frequently asked questions

Which IAB taxonomy do you return?

IAB Content Taxonomy 2.2 on classification, with category IDs, tiers and a score. Advertiser opportunities use IAB 3.1.

Do I need to upload files?

No. You send links to publicly accessible videos or pages.

Can you classify against our own taxonomy?

Yes, custom models are available on request.

Whose models, and at what scale

Our own models, trained and curated in-house by advertising and media people, so we can tune them to your needs. Your video isn't sent to a big LLM provider.

Built for libraries and feeds: our infrastructure processes 100k+ videos per hour.

New signals added regularly. Custom models on request.

Ready to test it on your own video? Get API access.