How is this "streaming" and how is it "live"?
It's neither streaming NOR live, by my definitions.
It's generating single images, one by one, based on a prompt. So it's automating image generation via calls to Gemini. That's it.
It's not generating "websites", it's generating an image based on a prompt, then generating a new image based on the previous image, and the position of where you clicked it, and a prompt.
The author has massively exagerated the complexity of this, you can one-shot this with a single prompt in Claude or Codex.