But it isn’t creating what I had in mind, or envisioned, if you will.
With AI systems you generate something with one action, allowing you much faster iteration loops. Remember, the author argues that the current prompting still has bad usability. Presumably a system with good usability could allow you to generate what you want with one, or a couple, of attempts.
There is no magic "paint the picture I want" button in Photoshop. There is a magic button in AI tools. This magic button does in one request what would take dozens or hundreds of commands without the magic button.
It very rarely does. And the more specific the request, the more work you have to do to get, at best, close to what you want. You may need to train your own model or blend models and mess with ControlNet. And then you have to do more work to fix stuff like fingers and eyes. In order to really use AI to its full potential (which still isn't, IMO, fully up to the capabilities of the software or artists it's intended to replace) you have to understand how it works, how the various filters work, how models work, etc.
Once you learn most of the really good AI generated art you find online is still heavily edited in Photoshop by actual artists you realize it's anything but magic.
