so, background, I’m currently workshopping a Disco Elysium/Touhou Project fusion fanfic (Disco concept, Gensokyo setting) and, since I already pay for NovelAI for recreational purposes, I figured I’d generate some character/skill portraits for it to accentuate the presentation of it.
Here's some of the better ones as of yet, alongside the one in the thumbnail.
However, I am conflicted on actually using them.
I figured I’d outsource the question because I’ve been thinking about it for a week and am no closer to resolving the internal contradiction. Here’s what I’ve been thinking thus far:
Concerns: I have moral concerns about the scraping and the ‘ask-for-forgiveness-not-permission’ philosophy that is central to these major generative models. I also think, in the context of it’s use within the hands of the bourgeoisie, AI is used as a tool of displacement and discipline. I also know that there is generally a negative perception of people who use AI image generation, for mostly valid reasons as most self-proclaimed ‘AI artists’ are coomers/techbros who make it a mission to discredit and disparage human artists. Even though I don’t fall into this category, I’m worried that I’d open up a vector of harassment where there otherwise wouldn’t be one if I include them, and I don’t want to deal with that.
Benefits: It adds to the vibes of the fic, and I think there’s a tangible enhancement with adding visuals to it. I also like what I’ve output so far.
Caveats: I don’t plan to make money off of this. I don’t have the disposable income to commission portraits for what ultimately is a fanwork that I don’t plan to profit off of and don’t have enough investment in to sink money into. The concerns are specifically what are holding me back in adding them, because it would be nice to add a visual component.
I just wanted to get a vibe check on here for this use case. Want to see if it will help me resolve the internal contradiction I’ve tangled myself in 
edit: I’ll err on the side of caution and drop the visual aspect of it, thanks for the feedback
Hard to compare, but roughly 0.005-0.05 kWh for an already-trained model to run once. A small number of additional outputs simultaneously can be produced for relatively little additional cost by most models. Edits and rerunning will vary, usually lower, but can rarely be more energy intensive.
For a digital artist to produce a rendition, it will vary a ton but for a relatively efficient PC with drawing software running, maybe 70 W, that’s 0.07kWh/h (or 70Wh) however you want to think of it. So multiply 0.07*(number of hours).
I think there’s stronger arguments against image generation, for commercial enterprises at least. I don’t really draw and haven’t inspected the files from image generators to know, but without recordings of the creation process I’d always wonder if someone I don’t particularly know that I’ve commissioned is really drawing it or not. This can only constrict us further, under capitalism. Fewer people will trust each other, just use machine models for low cost. Wealth probably ends up concentrating in fewer hands from there, even if more people are generating things, they’re vying for fewer patrons capable of compensating them. There’s some carrying capacity for entertainment, which is mostly what this stuff can produce at this point. Until it’s plugged into more robots, assuming things don’t break before that.