State of OME.TV

Nobody outside Ome.tv can answer that, and this report does not pretend to. What it counts is encounters: connections that members running the OME.UNDFND overlay were actually matched with, in a rolling 30 day window. That is a large continuous sample of the platform rather than a headcount of it, and every share on this page is a share of that sample.

In the current window: 3,136,252 encounters sampled and 3,746,706 addresses looked up, across 222 countries, by 4,047 active members.

The country table ranks where the strangers our members met were connecting from, resolved from the connection address. Read it as a picture of who our members get matched with, not as global market share: matching is influenced by where you are yourself, so a country being large here partly reflects where our own userbase sits. Countries with fewer than fifty encounters in the window are left out entirely.

Right now: Poland 58.4%, Germany 15.8%, United States 3.8%.

Traffic follows a daily curve with a clear night peak and a dead zone in the European morning, and a weekly one on top of it. The hour chart is published in UTC, because a single shared clock is the only way the number means the same thing to every reader, and it can be switched to your own timezone on the page.

Right now the busy window is 19:00 to 01:00 UTC, holding 51.7% of encounters, and Saturday is the busiest day of the week.

Most of them are over before they start. Skipping is the default behaviour on a random video chat, so the distribution is dominated by connections that end within seconds, and the interesting figure is the conditional one: of the conversations that survive the first fifteen seconds, how many make it to a minute. Durations are bucketed into four bands in the browser before anything is sent, so no exact session length is ever stored.

Right now 9.5% get past fifteen seconds, 2.4% past a minute and 0.4% past five.

The split published here is an automated estimate. A model running on the member's own machine reads sampled frames, and only the counts leave the browser. It describes the people our members were matched with, it carries the error rate of any face model, and it is published with its mean confidence rather than as a fact. No per-person record is kept and no image is uploaded.

Right now: 67% read male, 33% read female, at 91% mean confidence, with an average estimated age of 23.

Camera state is measured per sampled frame rather than per person: a frame is classed as dark, and the share is dark frames against all sampled frames. That covers people with the camera genuinely off as well as unlit rooms and covered lenses, which is why it is reported as frames rather than as users.

Right now 4.0% of sampled frames are dark, and a face is visible on 56.4% of the connections the model analysed.

What gets counted is addresses currently carrying a bot signal, not accounts and not people. A signal is raised by the detection network when the video on the other side looks like a recording rather than a live camera: a loop, a frozen frame, overlaid graphics, or the same face appearing from many unrelated addresses. One address can raise several signals, so the per-signal counts deliberately do not add up to the total.

Right now 11,898 addresses carry at least one signal, most often frozen frame.

The 30 day trend on this page is the honest answer to that, and it is republished every half hour. It counts lookups our members ran rather than the platform's own traffic, so treat it as a pulse rather than a user count: it moves with real activity, and the week on week comparison is calculated across complete days only, because the day in progress is always short.

Right now: 120,861 lookups on an average day, up 20% against the week before.

Members can correct it, and the correction is what the map is built from. A correction is filed against a network block rather than a single address, because addresses inside a block move between customers and the block does not. Nothing is served on one person saying so: a corrected place has to be agreed on independently by several members, with one place holding a clear majority of their votes, before it replaces what the commercial sources say. Blocks that do not clear that bar are marked contested and the panel shows the competing places with their shares instead of picking one.

To date: 915 blocks corrected by 320 members across 8 countries, of which 0 have cleared 5 votes at 60% agreement and are being served. 58 are still contested.

Yes, with a link back to this page so a reader can see the method and the window the figures came from. The same snapshot is available as JSON at /api/platform-report if you would rather read it as data. Please quote the window with the number: everything here is a rolling 30 days and it moves.