This isn't surprising, but is disappointing. Stable Diffusion 2.0 removed NSFW images from the training set making human anatomy generation a lot worse even for SFW prompts. This Techcrunch article is pretty unfairly biased against all AI art generation.
I dislike how much more political this debate has gotten. It's now impossible to defend one "side" without having a lot assumptions made about you, and often get anger aggressively targeted towards you. Very little online discourse seems to be happening in good faith.
There's a bunch of people exploiting the fine tuning to make models based on popular artists who explicitly don't want their art used in training sets. There are also a bunch of artists and non-researchers spreading misunderstandings about how art generation works and targeting researchers and AI artists very personally on places like Instagram and Twitter. I understand the anger people feel from both sides. But there's a very clear pattern of assuming people on each "side" as equally harmful as the most problematic on both "sides".
I think having a deeper understanding of the technology and it's limitations makes it clear that the best art is made by already trained artists. There's things that are generated wrong, like shadows in complex scenes, that everyone can recognize but only artists have the knowledge to recognize and fix properly. Like most things, the optimal solution is a hybrid approach, like using generated art as inspiration or mockups.
Midi instrument libraries exist but movie composers like Hans Zimmer still hire out philharmonic bands to record the score. I can't tell the difference but composers working at that level can. But they also have music banks of recorded music they rely on for certain instruments. But it's clear they will always prioritize live recording. Art is no different, in my opinion.
AI can generate an idea but it will not generate art in the exact way you want or imagine. Some situations allow you to get away with good enough if you don't look closely, but that's rarely the point of making art in the first place.
I dislike how much more political this debate has gotten. It's now impossible to defend one "side" without having a lot assumptions made about you, and often get anger aggressively targeted towards you. Very little online discourse seems to be happening in good faith.
There's a bunch of people exploiting the fine tuning to make models based on popular artists who explicitly don't want their art used in training sets. There are also a bunch of artists and non-researchers spreading misunderstandings about how art generation works and targeting researchers and AI artists very personally on places like Instagram and Twitter. I understand the anger people feel from both sides. But there's a very clear pattern of assuming people on each "side" as equally harmful as the most problematic on both "sides".
I think having a deeper understanding of the technology and it's limitations makes it clear that the best art is made by already trained artists. There's things that are generated wrong, like shadows in complex scenes, that everyone can recognize but only artists have the knowledge to recognize and fix properly. Like most things, the optimal solution is a hybrid approach, like using generated art as inspiration or mockups.
Midi instrument libraries exist but movie composers like Hans Zimmer still hire out philharmonic bands to record the score. I can't tell the difference but composers working at that level can. But they also have music banks of recorded music they rely on for certain instruments. But it's clear they will always prioritize live recording. Art is no different, in my opinion.
AI can generate an idea but it will not generate art in the exact way you want or imagine. Some situations allow you to get away with good enough if you don't look closely, but that's rarely the point of making art in the first place.