This is a pretty thorny topic so I'll try to carefully add something new to it...
I'm not familiar with Adobe Firefly in particular, but I take mrimer's word for it that he's done the appropriate research, and believes it to be minimally-harmful (with regards to environmental concerns, plagiarism concerns, moral hazard, etc.). This goes beyond merely accepting at face value any claims Adobe themselves makes about their product, of course.
However I'm still mildly disappointed to see the use of generative AI in the process (and thus its visible artifacts even in the final product which was further edited by human artists), for the following reasons:
1. The inconsistencies other people have pointed out (character design being inconsistent between scenes, lighting and focus being inconsistent, background characters not being as visible as they might need to be, awkward stances and angles). These are, of course, hard topics for human artists to get right too, but when they do get them right, it shows attention to detail. If people see an AI-generated image with these kinds of flaws in it, it raises the questions of "
where else are shortcuts being taken"
and "
were the artists paying attention to how their assets came out"
, whether or not they're necessarily fair questions to ask.
2. Using generative AI is an endorsement of that technology. Adobe Firefly in particular might be minimally-harmful, but there are several other AI applications out there that are not. In general I don't want to see this stuff normalized because it's difficult to research which applications are minimally-harmful and we don't want increased popularity and usage of the harmful ones.
3. It promotes a certain degree of homogeneousness and muddiness when it comes to character design. I think the main character Fenix's design, the techno-roaches, etc. look distinctive and fine, but I'm not a big fan of:
- the trailer's floating glowy squares during the hallway walks (a distraction that make it harder to see details in the hallway, I didn't think it was appropriate for sci-fi tunnel atmosphere. Usually I see that stuff when there's a big fire or a forge or somesuch.)
- the trailer's big demon ("
mech lord?"
) we only get to see a couple of seconds of (isn't really interacting with its environment, just waving its arms idly, no sense of scale for height or distance reference, not much to suggest that it's technological in nature or what villainous activities it's up to)
- the boxart's dark knight lurking in the background (hard to see, not much distinctive feature to it, would be hard to spot in a crowd. If I saw some other art of a dark knight how I would I know it's _this_ dark knight.)
I certainly think human artists could take these designs as a starting point and make improvements on them... but this is the trailer that just got shown to the whole public, which usually suggests a certain degree of "
release-readiness"
to these assets.
===
It might also help to include, in the Kickstater's "
use of AI"
section, some examples that illustrate the language used, "
[AI generated images] were revised, redrawn, reimagined, or entirely replaced by our artists for the final product."
, with pictures of the source images and the final images that were made by human artists - that might give a better impression of how much adaptation and refinement went into the final product rather than just using the gen AI image verbatim.
===
None of these would be deal-breakers when considering whether to buy this product, they're mostly just making me roll my eyes. I don't think it's a huge transgression but I also don't think it's a good look.
____________________________
Quick links to my stuff (in case you forgot where it was):
Click here to view the secret text