ML models in general are not great when tasked with rigid stuff. Stable diffusion can make a cool picture, but getting it to make exactly the picture you had in

ML models in general are not great when tasked with rigid stuff. Stable diffusion can make a cool picture, but getting it to make exactly the picture you had in mind is mostly impossible. This is actually a great use case in the sense that the requirements are flexible, it just has to be something empathetic and directionally related. That bar is so much lower than a trustworthy diagnosis or even a document lookup, either of which can be right or wrong. When what you need is "an example of x", this technology is great.