@Rahll Finally, I think there's ethical usage considerations. I personally don't prompt for "in the style of" unless the artist is public domain; I *also* wouldn't ask a human artist to mimic another artist's personal style without permission, cause that feels equally skeevy to me.
@Rahll I think the encoding vs transformation debate would also have to involve how easy it is to get the original data out. It's possible to get a seed for a random number generator that outputs an image similar to existing work, but the generator obviously doesn't include the work.
@Rahll IMO the main differentiation there is "can you get (most of) the original data back out?" That's made more difficult to determine because of how much prompting/randomness is part of image generation, but aside from some super-common things (the Getty Images watermark) you can't.
@Rahll I also *really* hope it becomes "artists using new tools to augment existing processes" rather than the objectively-shitty "just replace people with AI tools" idea.
Optimistically, the communities I'm part of tend to agree? Just the very vocal assholes being inflammatory...
@Rahll Yeah, my rationale there is that you can't take 2.3 billion images and get a 5 gigabyte AI model out of it without a pretty significant level of transformation going on. In terms of sheer data that's like 0.003% of any one image (if I did my math right). YMMV etc etc.
@Rahll But I do personally think there's a boundary between usage & creation in the same way that using Procreate isn't wrong because Apple products are made in sweatshops.
There's a blurry line there wrt support, of course - I haven't given money to Stability AI, so.../shrug?
@Rahll Yep, there's 100% some ethical issues that need to be worked out with web scraping that haven't really been a problem up until now. I'm not personally sure where the line between research & product is/should be, especially given that IMO the training process is transformative.
@Rahll I think AI art has a "skill floor" that's higher than other...media? Would that be the term? Anyways, it's easy to make something that looks okay at first glance with AI tools. Of course the more you develop taste the more time & effort you'll need to put in to reach that level.
@Rahll Sure, I agree that there's a lot of work that goes into making *good* 3D art. I think the process of making good AI art is similar - there's a ton of control features that are used by the people who want to make stuff that's better-than-average.
@glassbottommeg P5 does have a long running time (probably 70-100 hours, depending), but I think it has a better mix of story-to-grind than XC3. Most of the time spent is in cutscenes or daily life, not "grind out battles until you can tackle the next section"
@glassbottommeg The early game's fights were kinda lacking in terms of agency. I get that they didn't want to frontload the combat system with All The Mechanics (also narrative purpose) but I still definitely had to keep telling myself "this is hopefully going to get interesting later on."
@CalicoPikachu They showed us Empoleon and Torterra to remind us what good evolutions looked like...specifically to increase the psychic damage caused by Sprigatito's evo.