I write this from the perspective of a author and small publisher who produced works that required art. Back when I was filled with a lot more energy and oomph than I am now (and more free time, too) I had some pretty large projects that needed a decent look to them. I juggled this through a series of techniques, all of which are legitimate approaches when your base project funds are close to $0, and your expected return could also very well be $0:
1. It's who you know! Lean on family and friends for free art (or yourself if you are so lucky as to have enough talent to draw something that isn't gut-churning to an unknown audience)
2. Royalty Free art! There's a lot of it out there, from free old pieces that exist in the creative commons to art packages that require a small one-time payment and a credit in your work. I bought a bunch of these back in the day, and the main problem (at the time) was that a lot of other small publishers did the same, so there was a period where you could see a lot of royalty free art getting churned through various products. That problem seems to have diminished with time, however.
3. Minimize the need for art! This is not ideal, as people like seeing art to break up the text, and being able to demonstrate some of what you are writing about in a meaningful fashion is very handy, but on occasion "no art" is better than "really, really bad art." I have a range of sample books from back in the day where I can point to them and say, "Yeah, we never played this game because the art was a total non-seller to my group." An art free version would, in those cases, have been preferred.
There is also item 4, the nuclear option:
4. Buy some art from a real artist! This is the point at which you have decided your vanity project is worth some investment, even if you have no guarantee that it will ever lift off or return a profit. You may, at best, have the glorious satisfaction of pointing to an art piece and being happy that your book has such a cool cover or something similar Case in point: my Realms of Chirak book for D&D 4th edition (yeah, someday maybe I'll post a 5E update....some day....) by Simon Tranter was well worth it, even if I didn't end up making a return on the book good enough to warrant the investment. I would totally do it again, just because Simon is a great illustrator, and managed to capture exactly what I wanted for the cover.
Since the days of 2010-2012 when I did most of the self-publishing the market has changed in interesting ways. For example:
--the rise and fall of royalty free art packages; I think a modest number saturated the market for a while, and people eventually moved away from them;
--The rise of Deviant Art and other art spaces where new and aspiring artists could be contacted by aspiring self publishers for reasonably priced artwork; usually presumably followed by a "fall" as the arists become more popular and command higher prices;
--The rise of Kickstarter, where all prospective gaming projects seem to dwell these days, chiefly because if you go to Kickstarter and don't take that opportunity to factor in the actual cost of real art into your project then you are doing it wrong;
--The rise of higher demand for quality art in products. Let's face it, the market is now saturated with high quality content in terms of appearance; getting away with lower quality or lower cost art is much harder these days, and usually requires special branding (read: marketing yourself as OSR) or at minimum learning how to repackage poor/low quality art as being a deliberate cool style option, actually (see: everything related to Mork Borg, ever) by expanding your graphic design/layout skills;
--And, of course, AI generated art is now a thing. A very contentious and rocky thing that will readily lead to metaphorical fisticuffs for many people.
This is what has happened with a book I just picked up called Gammadark, an RPG expansion for the Shadowdark RPG. It uses a lot of AI generated art (using Midjourney) and the creator has expressed sentiment to the effect that he has no budget for art, and is hoping to build up a budget for the future. My thoughts on this are that he probably needs to be more cognizant of the market and how the prospective customer base will react to a product that is essentially 90% AI generated art. But....he's already produced the book, and while I find no issue with the game itself (it's a perfectly fine add-on to Shadowdark for a Gamma World-esque game experience), in studying the book I realize that there are several other reasons to criticize use of AI art in a project like this besides the "think of the artists" component (which is a perfectly valid criticism in and of itself). So, here are my observations:
AI Art is Still Derivative: the immediate item I noticed is that many illustrations in the Gammadark book look oddly familiar, but not in a "I have seen that illustration" sort of way, but rather, "I have seen other illustrations that Midjourney likely used to borrow ideas from," sort of way. The power armor and robots all look suspiciously like Warhammer 40K space marine armor, for example. The many mutants in the book often feel like they reflect some other source without specifically being that source. So as I look through this book, I find its art, by being typically derivative as most AI art is, lacks a specific identity that I can pinpoint. It doesn't even do Gamma World art well, because to be fair I suspect there's not enough Gamma World derived art out there for Midjourney to absorb and build a style from (and also, most Gamma World art historically can be traced back to a small handful of real artists, many of who had their art re-used from edition to edition).
AI Art is Hard to Replicate and Has A Stigma: Here's the conundrum: you have a vision for what power armor in the world of your game looks like. You input lots of descriptions into Midjourney until something pops out that is close enough to your vision to count. Now you have a single illustration that works. Later, you get better at your trade and maybe have some money to buy real art from an artist. Problem! Your vision of the power armor or mutant or whatever has a single image from an AI source to go on. The new artist likely doesn't want to replicate that look, because its solidified from an AI piece that might be borrowing or extrapolating from other real pieces of art; it fails from an AI replication perspective, at least for now, because AI art generators aren't good at consistency (though this problem may eventually disappear). You may even have a hard time finding an artists who will accept a commission from you because they know you used AI art before them. So now you have to find an artists who doesn't care, and be ready for their vision to deviate from what you have created using Midjourney.
AI Art Often Informs the Text Rather than Being Informed: Another problem is where you keep trying to get a certain look, then just give up and go with whatever is closest in vision. Maybe Midjourney and other AI art products are getting better and more precise, but I bet that a nontrivial number of pieces in Gammadark were actually written to reflect the illustration rather than the other way around. I bet a lot of art in the book is there with text because the art was close enough, and other items go without an illustration because nothing could be found that worked. In a way this is an old problem; using royalty free art packs will lead to the same issue, but usually you then leave your text alone and simply let the illustration fill space. In AI art, you may suffer the temptation of generating some AI inspired text to go with it. This, I can say as an author, is a big no-no. You will crash and burn if you allow AI generated text too much room in your product (which I interpret as any room; can you tell I am a writer and not an illustrator? =) )
All of these issues are hassles that I think you can solve by simply not relying on AI generated images, or maybe doing a small percentage of AI images. Some other books I bought for Shadowdark recently clearly have a couple pieces of AI art in them, but 90% of the text is human-generated art, with the one or two AI pieces standing out as a result; they feel less like the publisher was saving cash on them, and more like the specific pieces were chosen precisely because they did meet a core illustrative goal not otherwise achieved by the low-cost artists the rest of the product used.
Gammadark, alas, uses so much AI art, and at times its text reads suspiciously like it has some AI generated elements, that I think this is a bad sign for the publisher to start off this way. I am not done with my read-through, however, and will discuss in more depth in a Part II.....