I think one thing you and many other people misunderstand is that the image generation aspect of AI is a sideshow, both in use and in intent.

The ability to generate images from text based prompts is basically a side effect of the ability that they are actually spending billions on, which is object detection.