It really depends. There’s some good uses, but it requires careful consideration and understanding of what the technology can actually provide. And if for your use case there isnt anything, it’s just not what you should use.
Most if not all of the bigger companies that push it dont really try to use it for those purposes, but instead treat it as the next big thing that nobody quite understands, building mostly on hype. But smaller companies and open source initiatives indeed try to make the good uses more accessible and less objectionable.
There’s plenty of cases where people do nifty things that have positive outcomes. Researchers using it for pattern recognition, scambait chatbots, creative projects that try to make use of the characteristics of AI different from human creations, etc.
I like to keep an open mind as to what people come up with, rather than dismissing it outright when AI is involved. Although hailing it as an AI product is a red flag for me if thats all thats advertised.
You are probably confusing fine tuning with training. You can fine tune an existing model to produce more output in line with sample images, essentially embedding a default “style” into every thing it produces afterwards (Eg. LoRAs). That can be done with such a small image size, but it still requires the full model that was trained on likely billions of images.