Edge Video AI

Responsibility

The picture never leaves the TV.

We design our systems so that privacy does not depend on a promise. Edge ACR is our own content recognition on the TV, talking to the vision language model (VLM): the picture never leaves the set. In live streams, we describe the programme, never the person watching it.

A compact signal taken from the frame on a TV
Edge ACR on the TV. One read is 72 bytes. Frame from Edge's explainer film.

The picture never leaves the TV.

Edge ACR runs on the television and talks to the VLM. The picture stays on the set. A read is sent about every five seconds, and only while a listed programme is on.

Read the ACR report

Our commitments

Programmes, not people

Live enrichment describes what is on screen: categories, keywords, brands. It is not used to identify or profile viewers.

No naming of individuals

Our model is instructed not to name individuals in the content it reads, and to prefer the subject of a scene over incidental detail.

The stream can't give orders

Words spoken or shown on screen are treated as evidence, never as instructions to the model.

Sensitive topics handled with care

Sensitive topics are screened before signals are published, so ad decisions can treat them appropriately.

Fail open

If our answer isn't ready or isn't fresh, we send nothing, and our partners' systems carry on as they would without us.

Honest about what we know

We publish numbers with their scope, call demos demos, and never present brands seen on screen as our customers.

Questions

Talk to us

If you have a question about how our systems handle data, write to joe@edgevideo.ai. Our current privacy policy is at Privacy Policy.