How do you feel about AI art on Derpibooru?

I might be kicking a nest here, but I’ve been noticing a slow but steady increase in AI posts as the months have gone on, so this feels like a good time to ask how people are feeling about AI art right now, compared to when the “two per day” rule was introduced.

Are you more against it than before? Have you softened up to it?

Personally, I will not hide my bias; I’m am vehemently against AI art and am eagerly awaiting a solution from either the top (copyright legislation) or the bottom (Nightshade). All the meme stuff has been great fun, but in the end I can’t abide by AI art both because of the threat it poses to people’s livelihoods and reputations, and to the creative spirit of artistic creation and consumption.

To put a number to my own worries, I decided to count up how many AI images appeared on the first ten pages on Derpi, and ended up with this:

Pg 1 - 4
Pg 2 - 2
Pg 3 - 2
Pg 4 - 6
Pg 5 - 4
Pg 6 - 0
Pg 7 - 3
Pg 8 - 5
Pg 9 - 4
Pg 10 - 1

20 images per page,
31/200 images were AI-generated. Note that I only counted an image as AI if it had the AI tags.

So 15-something-% of the first ten pages were not real pony art. Granted, it’s a tiny sample size and there are probably much better ways of gathering this type of info, but we’ll call it a starting point.

So… do you consider this too many, too fast? Or are you just happy to see more pony?

@Parallel Black
Instead of manually counting a small sample size, you can use the search enginge here to easily get large sample sizes. For better data and for an actual comparison, I searched for total image uploads vs ai content uploads in january 2023 and december 2023.

January 2023
18498 total images
678 ai images
3.6% ai content

December 2023
19971 total images
1105 ai images
5.5% ai content

There does seem be a slight increase, although I think that’s likely due to ai generators being more accessible/available to normal users along with some of them getting slightly better with pony content over the year, getting more users to use and upload ai images. Keep in mind that as I only checked those 2 months, it is possible one of those months were an outlier either being higher or lower than the normal trend between months. I don’t feel like spending the time going through each month to see a more overall graph of how common ai uploads are, though if you really want to you could easily edit my search queries above to find it yourself.

As for my bias on this, I don’t like ai images either, although I don’t see much threat to them. It’s usually very easy to tell when something is fully ai generated, and I’ve only seen a few ai images that at least look ‘right’ to me. Although I don’t specifically search for them, I don’t filter them so I have seen plenty just around the site and when dealing with mod things. With all the imperfections of ai generated images, it really can take someone hours to get an ai image that matches exactly what they want and looks good. With those hours of work elsewhere and just paying an artist for your image, you get better quality without having to mess with any of the ai generators. So I don’t see any real threat towards artist commissions, at least with how the generators are in their current state. Along with the default filter here hiding fully ai generated images, and other sites like e621 fully banning ai content, there’s also plenty of encouragement against ai generated images for future use. I’ve seen worries that the generators can get even better but ai content seems to be reaching a ceiling in quality, new improvements are either slight manual tweaking or improvements to how fast they can generate.

So in short, there does seem to be a slight increase in ai content, but even as someone who doesn’t care for ai content, I see no bigger reason to worry now than before.

There are some high quality AI images (especially recently), which are amazing, and I even favorited some, because I loved what they represented.

I don’t like AI images which are low quality, has glitches, etc.

But sometimes I see something really good, and it deserves to be here.

Think of it as people having a way to realize their imagination and idea, even though they lack the skill. Why their imagination should be wasted and never be seen by others simply because they are unable to draw what they see in their minds?

Besides, creating a high quality AI art is still hard work in its own way, as it needs specific prompts (positive and negative) and can take time.

Don’t have a lot of compassion for it with how 99% of it I see here is stock solo female explicit content and all looks the same. Disappointing that the creativity bar is so low and people feel like they need to share all of the samey things they prompt for. There’s endless potential, and yet… Sigh.

@GlitchedWolf
Thank you! That’s the method I was missing. I’ll keep those tags in mind next time I perform a check.

In terms of the threat, small-scale, I’ve heard a lot of artists on Twitter saying how their commissions dropped off a cliff after AI became a thing. And large-scale, we’re already seeing companies reorganising their creative teams or downsizing to prioritise using AI either as a large part of the process or as a replacement for it. You gotta keep in mind that the average person is willing to settle for a lesser result so long as it’s quicker and more convenient. Food delivery vs home cooking, to bring the obligatory food analogy into this.

@Background Pony #7483
I would definitely dispute quality being any kind of difficulty these days. Putting aside the whole “you’re just typing words!” part of the process, most AIs seem geared towards giving very high-quality results as a default. If you’re talking about consistency, like less mangled hands and stuff, then… I’m not sure about that, either. No matter how many words you enter into a prompt, the actual process is still little more than a slot machine. You can base a result on an input image, too, but even then the results will be a spread of possibilities where you simply select your favourite. At the end it’s unlikely to be what your idea would have turned out like had you had the skills to draw it. Even if one’s skills are entry-level, the drawing process still plays a big, big part of bringing an idea to life.

@Background Pony #933F
This plays into what I was saying about people being willing to settle for less. They take what the AI offers them, which ends up being the most generic-looking stuff.

In my ~10 years of using this site, the first and only type of images uploaded that ever convinced me to permanently switch away from the “Everything” filter is AI. I want to avoid making a value judgement here on the content of the images themselves, because there are rarer cases where it’s done well enough that it isn’t unpleasant to for me to observe. The filter system is operating as it’s designed here, but I feel I can’t be alone in thinking this type of imagery is not a great fit for an art board.

ed:

This is simply not a sustainable solution because of the nature of generative adversarial networks - they will simply add the techniques implemented by Nightshade and Glaze into their adversary function.

I ranted about AI some time ago so I’ll be brief. AI art is stupid, but it clearly isn’t going away and the AI situation will only get worse. It’s easier and more convenient to just use AI to create art and while it’s still limited now, eventually AI generated art will be of higher quality than even the best art of the most talented artists, and I feel like at some point, almost everything here will be AI unless Derpibooru staff does something about it or the website shuts down for some reason before that can happen. AI hasn’t completely replaced human artists yet, but when it does (and it will), I’m going to miss when artists actually drew/painted/created their art instead of having machines do everything for them.

Yeah that and foalcon for me is why I avoid that “Everything” filter also. I did make a public “Almost Everything” filter that shows everything except AI content and a couple other things.

I remember back ages ago, I was very much for banning Gmod/SFM from the site, as well as “based used” stuff. The vast majority of Gmod/SFM back in the day was trash anyone could make just by loading up some premade assets and moving some gizmos around for a couple minutes. Not much to say about “base used”, plop a premade PNG and use the fill tool for a minute, bam, new “art”.

Fast forward years later and Gmod hasn’t visually improved much but is at least used for memes or comics with a story. SFM (and now blender) stuff has gotten much better and people are putting in a fair amount of time and effort into it, with many SFM/blender artists eventually making their own models and animations. “Base used” still exists, but if you track people that use it, a fair number of them eventually draw their own new things without bases, or at least heavily build on them.

Right now with AI, the tools are still changing month by month, people are seeing the latest version of this or that model/generator/technical doodad coming out and giving it a try. There are a handful of people that have stuck with it, but most are either scratching an itch to see their waifu/fetish/fantasy, or giving “the thing in the news” a go and then dropping it (until the next time it’s in the news again).

My expectation continues to be that most people using these things will move on after “getting it out of their system”, a few will stick with it because they genuinely enjoy and like it, and some will become “mixed media” artists combining “traditional tools” (like photoshop…) with AI. Either way, it’s been a year and the tech is still constantly changing, so it’s hard to truly judge.

@The Smiling Pony
My expectation is that most people will keep using AI because it’s easier for them and it’ll be more convenient (the same reasons they move to a lot of new things) and almost everyone else will eventually give in and go along with it because they have no choice.

@Starless
I can tell you that AI does not make anything close to what I can create and it is also immensely terrible at OCs. Even if that changes in the future, it’s not gonna make me ‘go along with it’.

Its creepy.

Honestly I’m ticked off people are still calling it AI. It should be ML, for Machine Learning. There is no intelligence involved in anything these programs are doing.

But aside from that, definitely more against it. And every instance I see of it in use, or promotion of all the things it supposedly can help with, increases the irritation I have with it.

It’s a waste of server space and site resources and Derpi should stop enabling it.

@Background Pony #D7C6
Its not really learning either as it more like random image generator.

@phoenixacezero
Yup, not ML so much as random statistical emulation of other people’s art, without permissions.

January 2024: 8.44%

@ZippySqrl
Oh. Assuming all the numbers are accurate, that’s a rather dramatic increase compared to December…

@Parallel Black
Yeah and it’s only going to get worse.

February 2024 (so far, way too early but still): 8.95%

@Starless
I knew I wasn’t imagining a sudden increase. That sucks.

Here’s the current AI upload rate from the past ~12 months

it has definitely increased recently.

@ZippySqrl
Interesting. Is this a graph you made, or can it be found on a site somewhere?

@Parallel Black
Made a quick little program to extract and collate the data via Derpi’s API, then slapped that data into a Googledoc Spreadsheet to create the graph.

@ZippySqrl
I see! It looks to be the perfect way of keeping track of all this, so good stuff.

AI images are not art.

So it looks like videos are going to be next on the “actual effort/creativity versus convenience/immediate payoff” chopping block. OpenAI has released a demo showing off the capabilities of their video generation software, Sora, and I doubt it’ll be much longer before before an endless stream of samey looking garbage animation starts clogging up the uploads here as well.