It's not weird. One of them is piracy to play games, watch movies, and post throwaway memes. The other is for training generative models that directly compete with the creators of the pirated works and drop the value of human creative skills to zero. The actual act of piracy is small compared to attempting to devalue and commodify our most human characteristics and abilities.
AI is very good at low effort content. But that's not really competitive in the professional market. There it's just a nice tool that still needs an user with artistic skills.
AI competitiveness depends very strongly on whether AI will get better at creating coherent and usable video/audio/images, or not. The better it gets, and the more amenable to user control it gets in the fine details, the more it will displace actual artists.
And I don't know the answer to that because it's not clear what the limits are on current techniques or whether anyone's working on something that would transcend the current techniques. I agree that current techs are not really suitable for high-value content without human followup; SORA might be lauded but it still does not understand what a human body is and will happily give it a random number of limbs in various stages of simultaneous movement.
(But, for the legal issues, I believe intent to compete is enough and not just actual competitiveness. E.g. fanfic is not actually legal unless clearly a parody or meta-commentary or the like, but it's ignored because it doesn't directly do anything negative to the media market.)
I've seen a D&D subreddit that bans AI art but encourages the use of art taken from google images.