Raw LLM Responses

There is probably also a huge conflict of incentive in how AI is fundamentally designed to create images. Like with LLMs, where they'll lie if they don't know the answer, because training rewarded that, why should an image generator that is tasked to "Create Homer Simpson holding a bowl of rice" create an original, inspired homer instead of exactly how homer looks in the tv series. Why would the prompter want to have the AI put their own spin on homer, if they want a copy. And if you'd prompt "Create an original image inspired by Homer Simpson holding a bowl of rice", you'll probably get something that the prompter doesn't like and will make him waste time re-prompting a lot (wasting money and/or time). So I'd say that there is literally no incentive for AI models to be trained to be an original artist that has their model-specific style. Nobody would use it, but it would be closer to what an artist is. Artists are not photo-copiers, but in order for any single AI model to satisfy the masses, those models have to be (and no, specialized models are just better at a certain category due to more training material for that, they're not more unique). The best analogy is probably that an image generator is akin to a very good person who traces art they like and try to pass it off as their own. They're also shunned. AI models are literally the same, just that most people who do trace art do so to learn initially (which the AI kinda did as well during training). But the prompter or model will never become a unique artist. They'll just prompt away, best case just add a similar set of keywords to each prompt, so images appear stylized because they always combine the prompt with a different, always same thing.