Sloppacurricular - Sunday Slop Edition and Thoughts on AI

With Flux launching I decided to fuck around and generate some "realistic" versions of the Extracurricular girls before the inevitable prompt filtering happened to dissuade horny slop.

A lot of people have asked me about AI generally, since it's a hot topic in my particular trade. While I'm not as aggressively hostile towards it as a lot of my peers, I also don't really think it has much (if any) creative merit and extremely limited potential for quality, practical applications, regardless of what advocates proclaim.

First up, the lack of creative merit. Tech bros will say that "prompting" is an artistic skill in and of itself. That it takes effort and "talent". This is complete nonsense. You just type shit. Typing shit in a certain order weights things differently. That's it. You type shit, you pull the lever on the slot machine 50-100 times and you pick out the 1 or 2 results that are sort of close to what you wanted.

Second, why I feel it's not particularly useful outside of being a toy or a dopamine dispenser (one that uses eye watering amounts of electricity). The only actually worthwhile use case is just using it instead of stock images, but stock images exist already, have done for decades and do their job perfectly. Photographers who specialise in stock images are probably the creative arts professionals most at risk from AI ruining their trade and have a right to be worried and hostile. So why is it useless? Essentially, continuity. It can generate some quite impressive one off images, but struggles massively with cohesive, repeated detail across multiple images due to the fundamental way in which it operates. There are attempts to rememdy this, such as inpainting, referencing source images, algorithms that filter through hundreds of results to find closest matches etc, but they'll never be perfect, again due to the fundamental limitations of the diffusion approach. None of these technical limitations even take the sterile "soulless slop" factor into consideration.

In summary, I don't particularly like or hate generative AI and I'm not worried about it taking my job. People have sent me multiple instances of attempts to train SD models on my own work and the results are always absolute dogshit.

Thanks for reading my essay. Now my thoughts on my own "efforts" above.

----------

I think Teagan, Sunbeam and (especially) Erin best capture my in-brain envisioning of the characters. Erica is terrible and the rest are just ok.

As a "bonus" I ran them through Luma (I can't believe people pay for "no watermark" versions of this shit, it barely works) with uh, varied results. Again, Erin came out the best. Nathalie's face morphs into an entirely different person. Sunbeam's fingers turn into hotdogs. I omitted the results where Stephanie turns into a skateboard like some sort of cursed biological Transformer.