Every frame and every second of sound on KEEP WATCHING is made by an AI model. None of it was filmed. There was no camera, no microphone, no location, no crew and no performer.
The places do not exist. Anything resembling a real person is not one. Anything that looks like a real event did not happen. Nothing here is documentary, reportage, or a record of anything.
Each clip starts as a written description of a scene — the look, a timeline of what happens second by second, one camera move, and a direction for the soundscape. A video model reads that and produces the picture and the stereo audio together, in a single pass.
That last part is the unusual bit and it is why this sounds different from most generated video. The sound is not library music laid over the top afterwards; it is produced from the same description as the image, so the rain you hear is the rain you see.
No accounts, no sign-in, no advertising, and no cookies. The live viewer count works by your browser sending a random number it invents on page load, which is forgotten a minute after you leave. It is not linked to you, and there is nothing to link it to.
Because generated video is mostly served as an endless scroll that needs your thumb to keep working. This is the other thing you can do with it: put it on a channel, give it sound, and let someone leave it on.