Using an AI art generator to make graphics

Authoring General Design Discussions
,
61

Open question: if you have a ethical/moral/legal objection to generative AI, does that mean generative AI in general, or is it specifically because of the use of non-public domain training data?

So, for example, would there be the same objections to a game that used art generated with Stable Diffusion using one of the models trained entirely on public domain data? Or a a purpose-built model using a specific corpus of public domain data? Or a model trained entirely on one of the authors’ own art?

Not trying to post this as a gotcha or anything, honestly just trying to take the temperature of community sentiment here.

5 Likes
62

Interested in people’s answers to this one. I’ve seen a few “ethical AI art” accounts on Twitter that use open-sourced AI models trained on public domain images and the like, and charge commission prices analogous to the prices a working artist would charge to not devalue the effort that goes into making an art piece. (If I remember right. It’s been a while.) Haven’t looked into them too closely and I’m a bit skeptical, but it’s an interesting concept. Corporate AIs will probably win out in the end, though. May be personal pessimism but it’s hard for me to see ethical AI gaining traction when companies like OpenAI already have so much money and VC support behind them.

Don’t want to say too much in this thread since AI art is a touchy topic. And as someone whose livelihood isn’t affected by it I don’t think I have the right to go on about its pros or cons. I’ve had fun with DALLE2 and other AI generators, and used them a little, but after seeing the ethical arguments and how opposed many artists are to it, I probably won’t use AI art in anything else I make. Personally I don’t mind if others use AI art, though, and it won’t put me off a game as long as the game’s good on its own merits. That said, I think that barring experimental works, the point of interactive fiction should be writing, not visuals, and good writing produces its own imagery. If a story uses AI-generated visuals, it might be better off not having visuals in the first place.

7 Likes
63

I can understand this point of view, but for me, visuals can add to the atmosphere and mood of the game in a really fun way. There are games I love with no visuals, and games I love BECAUSE of the visuals. I like text adventures because I like the puzzle style, not because I like reading IF.

I’m sure very few people will relate to this, but: Nearly every game, text adventure or otherwise, that I’ve made, I made the graphics before I did anything else. Sometimes I made them before I even knew what my game was about. (I mention this in the blog post.) I use the graphics to inspire me. Sometimes, an idea starts to form as I make the graphics, other times, I look at the images and think: what could be going on here? So, for me, they are pretty important.

10 Likes
64

You may not want to read my take on this if you have no heart at all for Luddite ideas.

Are you sure it’s not just that you don’t want to forego the frivolous amenities of modern life? It’s possible to downscale one’s life to pre-industrial levels. After all, the downtrodden poor-folk of former times made do, too: They walked barefoot, except on Sundays, their clothing was patched and laundry was done only for undergarments, once a week. They ate oatmeal porridge and seasonal vegetables. Fruit was a treat, meat was reserved for solemnities. They repaired their tools, their clothes and their housing themselves, and often produced many of these things by themselves in the first place. Education was limited to the bare basics, because the productivity of young people could not be wasted on the study of books. Until quite recently, almost everything the common people consumed or could reasonably aspire to was made by hand.

And yet, it is under such circumstances that most of our current ethical standards developed. I believe the current fashion of putting down the efforts of former times in this regard - for example because they exerted much pressure on the individual to conform to expectations - is misplaced. They maintained most of the core moral and cultural program underlying society to this day under very trying conditions, and that may well have made them more ethically correct than we are today, for it could be argued that in order to preserve the planet’s natural resources for future generations, no one should be allowed to consume more than the product of their own hands’ labour. You want help from a machine? Build it yourself, in your own time, but don’t use more than your fair share of materials, please. That’s ethical, and I’m afraid we may not be able - nor were we really ever able - to afford more than that.

This is emphatically not the same as living a hermit’s life in a cave, a pious abstinence from anything remotely fun from cradle to grave. On the contrary, a rich life comes from all the good things we do, if we dare to do them: song, dance, music, theatre, play, sports, art, literature… we don’t need an AI for any of this.

5 Likes
65

Society was built around different expectations in the past and I bear no judgment towards people that want to grow their own food, prepare meals from scratch, make their own furniture and clothes, etc. For the majority, I suspect work and the expectations of modern society would not make that an achievable goal for many, though (I think if I turned up for work in home-made clothes, I would quickly find myself out of a job).

But this isn’t really the point of this thread. Is buying a pre-fab wardrobe from IKEA unethical because it diminishes the value of the carpenter’s skill, for example. Well, that’s nothing to do with IF.

The question posited in this thread is to discuss the impact of using automation to enhance the presentation of a non-commercial niche hobby project like a free interactive text adventure game. The thread keeps turning back to wider societal impact of AI (which we’re never going to address in this thread) without much actual discussion about the impact of using AI tools enhancing IF (predominantly images, but text and sound are also relevant to this thread, I would suggest).

5 Likes
66

With all ethical constraints removed, I reexamine my priorities.
It would appear possible to have the AI illustrate a game lavishly by the standards of IF. If you have a good generative AI, know how to prompt it and take the time to weed out and improve the imagery, you can have pictures for every room, its scenery, all your interactive objects and NPCs, possibly including different expressions and body language at comparatively little effort. Similarly, you could, of course, have an AI compose a score.

Creating pretty sizeable visual novels for example would no longer require much of a team effort. But these possibilities could find their way into other forms as well, thereby bringing, for example, new HTML choice games closer to VNs on average after some time.

There will be those who constrain their work to make the most of the generated content (i.e. fewer, but better illustrated stuff) and those who never finish their games, failing to live up to their own expectations regarding visual and auditory polish, even though they could have published their magnum opus text-only. That’s not new in principle, though.

Generating the text itself is another point (“Write me a parser-based detective game in the style of Dashiell Hammett. It should include the player fixing the murder of two corrupt police officers to look like self-defence and framing a known group of gangsters for what was really just a tragic accident in order to raise the public’s wrath and cleanse the town of the criminal element.”). No one need to worry about being able to write memorable descriptions or snappy dialogue ever more. But we wanted to stay on topic, which was illustrations.

67

No need to remove all ethical considerations, just limit them to the scope of IF. Do people feel that generative AI will have a positive or negative impact on the quality of IF, or on the accessibility of IF, or on the attractiveness of IF to a new audience, for example. Does the use of generative AI disadvantage writers or illustrators of IF that do not augment their craft with clever software?

It’s obvious from the contributions to this thread that it is a truly emotive topic, so it is definitely worth exploring. This is something that IF needs to wrestle with, after all - AI is here and people are using it - so how can we, as IF players/authors, use generative AI responsibly and fairly in our projects? And I’m not really talking about commercial projects (because I suspect people profiting from AI changes the considerations), I’d prefer to explore the use of generative AI in hobby projects like the IF most of us write/play.

2 Likes
68

But the ethical considerations as regards the use of generative AI in the context of IF in particular are not so different from those in the general context: Are we fine with reducing our role to consumers of computer-generated content or do we consider the dignity of productive work to have value - for us personally, as regards making interactive fiction - as well?

Of course a bit of experimentation with the possibilities of AI, just for fun, just out of curiosity, just for ourselves, without any plans to publish the result, won’t hurt, will it? It’s not for profit. It’s surely not a frivolous use of valuable scientific equipment, of energy and our co2-budget. There has got to be room for some private experimentation. And where can be the objection to sharing the results of our experiments, for example, at the IFComp? It looks pretty awesome, we’ve got to tell the world, they’ll be amazed. And anyway, it looks like it was done by a human! So let’s see if they notice by themselves. Some of the other authors will probably have got to this point already. And even if they haven’t, they will be there next year, sure enough. And even if they are not, they would if they could. There is everything to be gained and nothing to be lost by being the first. It’s sink or swim.

1 Like
69

This seems to be where the argument keeps circling back to - all automation is bad becuase it devalues productive work. It’s the argument made every time in history that there’s been an advance in automation. Although I agree that there does seem to be something more emotive about automating creative work (creative work feels more uniquely human, but I’m not sure if that’s just sentimentality). If I use Adventuron to write an IF, is that undermining the productive work that a coder writing a custom parser could have been doing? Should all IF be created using home-made parsers? Or will using tools to accelerate the production of art and text be considered no different from using platforms and parsers? If i can visualise a picture but can’t draw it without help from clever software, is that any different from writing IF that I can’t code without the help of a parser engine? Does creative work have more dignity than coding work?

Given that most IF are solo projects, is it reasonable to expect that every IF author has the skills to write, illustrate, and code their adventure? Or is that barrier too high to allow new people to join our community?

7 Likes
70

Speaking for myself, I do not require anyone to reinvent the wheel or, indeed, to refrain from using technology x or y.

Nonetheless, I believe that people instinctively value products higher, when the use of prefabricated ingredients is lower. The art market is an example: The same picture, painted in hand-picked, hand-ground gemstone pigments suspended in egg-yolk will likely fetch a much higher price than if it was painted with acrylic paints from the hardware store. Illuminated manuscripts, although long since obsolete, trade for astronomical sums, even though etchings are so much more precise and the contents of printed books is so much more relatable. But even very nice and old printed books are cheap, by comparison.

Hence my expectation that some authors might end up wanting to keep their use of AI very close to their own little heart, yes, yes.

1 Like
71

My bad, I hadn’t realised your point was just about authors passing off AI-generated content as their own work. In which case, I’m in 100% agreement and I’d be surprised by (but would still welcome) anyone’s counter-argument to that position. I also agree that most people would attribute a higher value to something hand made than to something prefabricated, which could put IF authors that used generative AI under pressure to hide their use of generative AI. We see this all the time in photoshopped models in magazines and on Instagram and in the secrecy around steroid use. Pressure to earn money from (or remain ‘popular’ because of) one’s appearance puts pressure on those people to take edit their appearance or otherwise act dishonestly about their physiques.

So if that’s your objection to using generative AI in IF, if an author hides their use of generative AI in IF, does that reflect on the use of generative AI in IF or on that author?

3 Likes
72

lol

1 Like
73

There are clear-cut cases for automation: It used to be the case that the binmen had to lift the dustbins into the refuse lorries by hand. When someone invented a hydraulic mechanism to do this, everyone was relieved that these people had to do a bit less back-breaking, hazardous and dirty work. As you’ve noted, it’s perceived differently with artistic endeavours.

Where does your and my rejection of the unattributed use of automation come from? While probably relatively common, this sentiment is at the same time doubtlessly absent in large parts of the population. My guess is that these would be the same people who cannot recognise kitsch for what it is. Bear with me here for a moment.

It’s not so easy to verbalise the difference between kitsch and art. Commonly words such as “appeal to sentiment”,“effortless identification of and with”, “ease of consumption” and “derivative” will crop up, but there will be objections to all of these. Nonetheless, to anyone with even just a passing interest in art, the difference is usually obvious at the first glance, even though kitsch is not the same as bad art, and can exhibit quality workmanship. But in fact, kitsch is not judged along such a scale. To get there, a work has to communicate something of value. And there can be no doubt that this kitsch doesn’t achieve, because it explicitly isn’t meant to do so. If the negative artistic value of a low-quality painting of the Eiffel Tower is just the profit of the postcard seller, it’s probably tolerable. If the object of some gigantic, academy-quality, horned-helmet-nordic-fantasy history painting is the indoctrination of a country’s youth to go to war with their neighbours, it’s worse. But my point remains in both cases: There can be no fruitful discussion about the quality of kitsch.
How do we spot kitsch so easily? One way to go about it is this train of thought: “This is clearly derivative, but people who haven’t seen a lot of art won’t know it. It has been created to appeal to such people specifically. Now let’s ask ourselves with which miserable intention this was done.”

AI generated imagery is, by definition, derivative. It comes from broad consensus and not from the discovery of underappreciated gems. And I think the common negative judgement of the derivative nature of kitsch comes from its closeness to plagiarism.

For comparison: At least until recently, people just the slightest bit read would easily spot the typical, insufferably soft-spoken, anxiously balanced school-essay style of ChatGPT, in the same way as we spot kitsch.

To finally get to my actual point: if we are to judge the AI-generated content on similar terms as we would judge other works, we would have to talk about the moral value first, before we get to the quality of presentation. And, as you point out, we should not be talking about the moral fibre of individual authors. Instead - at least in my opinion - the ethical judgement on the use of AI will point to the temptation it offers for people to henceforth assume the consumer role and the possibility space it opens for the manipulation of ever larger swathes of the population, because one thing is certain: it will only become harder to spot the difference, not harder to come up with nefarious goals.

1 Like
74

That was a lot for someone like me unfamiliar with art terms to parse, but let me see if I can summarise - because generative AI can be used by malicious authors to manipulate people for nefarious ends, if we use AI for non-commericial projects like IF, this is contributing to the exploitation of the people malicious actors will manipulate. In the same way as, for example, people that buy technology or clothing manufactured in countries where workers have lower standards than we have for our workers in the Western world are contributing to the exploitation of those workers. So by playing IF on an Apple phone, the act of playing IF is exploitative of workers in Apple factories that have worse working conditions than workers living in the US.

Please correct me if I’ve misunderstood as there was a lot going on in the middle of your post about kitschness that I didn’t really follow. I think you were basically saying that AI doesn’t produce original “art”, it can only emulate the styles of existing artists (like how most human artists do - I suspect very few artists have an original style). Computers can draw an original picture of the Eiffel tower (i.e., it won’t be an image of the Eiffel Tower lifted from another piece of art, but the engine learns what the Eiffel Tower looks like - proportions, reflectivity, colour, etc., from lots of examples, and then constructs an original image from that understanding), but it can only do it in a cartoon style, or a Van Gogh style, or anime style, or photorealistic style, etc. It won’t be able to create an original picture in an original style (which I don’t see as a problem for IF, personally, because most IF art I have seen follows existing stylistic conventions).

1 Like
75

Sorry to derail the thread, but I just wanted to pop in to say thanks to @Grizel for the blog posts, it’s really interesting to learn about other IF makers’ processes. :slight_smile:

5 Likes
76

To put it a bit more bluntly: We don’t want to be fooled by someone passing off AI-generated content as their own work, because not only do we feel cheated out of our judgement on the effort it took to create it, we also feel (almost instinctively) that people making use of AI are usually up to no good, because for starters, they are not (much) above plagiarism, and from then on thinking about their possible motives only gets more uncomfortable.

2 Likes
77

Thanks for clarifying. So your objections to the use of AI in IF are:

  1. to the unattributed use of AI (which seems fair - the use of any unattributed source of art, text, IDE, etc., seems at best discourteous and at worst dishonest); and
  2. that the association some people make between generative AI and the malicious actors that exploit this technology to manipulate others would sully the reputation of IF to those people (even IF that does not use generative AI in its creation), thereby costing our community valuable new members.
1 Like
78

I didn’t think about that. I only tried to verbalise what I myself feel when I think about AI-created “assets” and that I believe that this sentiment is shared by some other people.
What I can imagine will happen is that in the absence of evidence to the contrary, we will have to assume every title will include at least some AI-generated content and therefore forego any special sympathies we might have held for work done entirely by hand. This I consider a loss.

1 Like
79

That’s a really interesting point (well, 2 points, really):

  1. The availability of generative AI creates the assumption by players that all authors will use it, whether it has been attributed or not. I guess there’s nothing we can do about that because generative AI is out there. Whether we use it or not, we’ll be judged against standards set by generative AI, but this does put a lot of pressure on authors to use generative AI.
  2. And if we use generative AI in IF, this will raise the standards of IF to the point that new authors not prepared to use generative AI will be unable to ‘compete’.

Pretty much damned if we do and damned if we don’t!

2 Likes
80

Hm, just to make sure: Do you mean this sentence in general?