Generative AI images on TBT

[Mistreil]

Hi everyone! This post is on the topic of Generative AI ("GenAI") images on the forum. I'll be going over the topic in quite a bit of detail to give context and explain the rationale behind some of the decisions made.

However, as this post will be long, a summary of the main 'decisions' that have been made:

• GenAI images are not allowed in The Museum.
• GenAI images are not allowed in events.
• GenAI images are currently discouraged but not disallowed in other places (avatars, signatures, and other boards).

Additionally, while this is not GenAI-related it is 'art permission'-related: if you have created art that someone is using, you may request for it to be removed (e.g. if proper credit is not given, or if it's being used without your permission.)

Please read below for more details, and please feel free to provide your feedback!

Context: What is the problem with GenAI images?
This section is a very simplified explanation for what GenAI images are, and a few reasons why they're disliked/controversial.

* As a disclaimer: I am very aware of how GenAI works at a deeper level than what I'll describe here! There's a lot of nuances that I am skipping over and simplifying for the sake of brevity. I have a graduate degree in computer science, which directly influences my understanding of the topic and my dislike for it.

What are GenAI images and how are they made?
Generative AI images are, as the name implies, images generated artificially. The specifics on how they're made varies, but the core parts which are relevant to this post remain the same: images are taken as training data and used to train a model, which can be given a prompt, and outputs a new image based on the data given.

As such GenAI can only produce things that they've been trained on: this mean that, even though the result might be new, it's still a combination of things that it's seen previously and learned.

Why do people dislike GenAI images?
There are a variety of reasons, but I'll largely be focusing on ethical concerns.

To start, as I mentioned before: GenAI models are trained on existing images. The images used to train these models are usually used without permission: for example, taking images from google images, artist portfolios, social media, etc.

This means that the images produced are essentially products of art theft, as its training data wasn't ethically obtained. Often, people are able to give prompts asking for art drawn in certain styles and get reasonable results. This has affected both corporations (with Disney and Universal recently suing for copyright infringement) as well as many individual independent artists.

For individual artists in particular, I've seen the following happen multiples times:

• Artists who had models trained on their art being accused of using genAI (when, in reality, the art style resembles theirs because it was trained on theirs without permission).
• Targeted attacks towards artists done by training models on their art (and in some cases, posing as them for various reasons such as financial gain or to damage their reputation).

These are just a few instances of direct harm caused to artists, as I'm trying to focus on direct impacts for the sake of brevity and as I feel these get the main ethical concerns across more clearly. However, there's also financial damages (e.g. replacement of artist jobs due to the use of GenAI), as well as other discussions that could be had such as "what qualifies as art" and the commodification/cheapening of human-created art as a result of GenAI.

There are many reasons beyond ethical ones, and I've only really focused on GenAI in the context of generating images: other domains have other issues (e.g. deepfake videos being used for misinformation, as well as the consequences of depending on GenAI both for social reasons and educational ones).

While I won't delve into this topic too much, there are also environmental reasons for disliking GenAI. Training models takes a very large amount of power as the model processes many images, and running prompts themselves also take a fair amount of energy. There is a non-negligible amount of energy consumed: if you're curious, I do recommend looking into literature on the environmental costs of GenAI.
(However, even if the environmental impact is reduced by future technological advancements, the ethical issues will continue to persist.)

The Museum and Events
GenAI images have been and will continue to be strictly forbidden from being posted in The Museum and in events. GenAI images are not art, and are more in-line with art theft, which we do not condone nor wish to encourage in these spaces whose purpose have been for creations made by members in order to showcase their abilities and efforts. Trying to pass off GenAI work as your own is strictly forbidden and may warrant stricter punishments depending on severity.

GenAI images as reference images for event submissions
While we strictly forbid GenAI use in events, we do wish to bring up a slight gray area. Though we allow referencing of images as long as credit is provided (e.g. in the form of posting references used, mood boards, etc.), there is a bit of a gray area in the case of "what if a submission is made without GenAI, but one of the references used was a GenAI image?"

We will be evaluating these on a case-by-case basis: the severity of the usage, whether it was intentional or accidental, and the member's history will all be taken into consideration. Largely, we do not wish the penalize members who may have accidentally used a GenAI piece in their moodboard or as a reference, but we do not wish to let it slide for members who purposefully did so.

In cases where we cannot confirm the use of GenAI in this manner but hold strong suspicions, we will likely disqualify the submission from winning a larger reward (e.g. staff favourites) and keep an eye on the member for future occurrences.

If you do see suspicious entries, please report them! Sometimes we do accept entries that should not have been accepted, as some staff tend to accept things in batches and may miss some things.

Discouraged, but not disallowed in other places
While we understand that GenAI images are disliked by many members for the aforementioned reasons (see Context), we will not be disallowing them from being posted in other places for now.

At present, GenAI images have not been used enough for a firm decision to be made, and we feel that forbidding entirely could potentially be too harsh. If an image was posted to contribute to a discussion, it's hard to judge whether this is inherently harmful or not without having any further context. Below are 2 examples that we believe are likely scenarios, as well as the rationale behind the lack of moderation action:

Example 1: Suppose there are discussions about a game and its concept art is posted, some of which end up having been made with GenAI.
While this constitutes as "posting GenAI images", this scenario doesn't seem like it warrants removal or further action, especially as the GenAI images were not created by the poster themselves. If anything, we foresee discussions on such games being centered around disapproval of its use and being educational in this way.

Example 2: Suppose there is a thread about a concept (e.g. a new villager species), and someone posts a GenAI concept image.
As long as the member isn't claiming that the concept image was theirs, this feels somewhat similar to "posting an image from a search engine". Additionally, with how search engines have been flooded with GenAI art, it's also become more likely that people will accidentally share GenAI images. If the intent wasn't harmful and was to further a discussion, then it's unclear whether such a post warrants removal or further action.

In trying to decide whether these scenarios warrant moderation, it's become difficult to determine whether there is a difference between a member generating an image themselves vs. posting one they found elsewhere. Similarly, if there is a difference it seems difficult for us to fairly differentiate between these without looking at them on a case-by-case basis.

At present, we do not have a metric for what may be harmful. As time goes on, we may revise our stance if posts with GenAI images are made.

With that said: please do note that if you are sending GenAI images to members that are uncomfortable with it and have asked you to stop, this would be considered rude behaviour and is punishable.

Aside: Avatars and Signatures
The use of GenAI images in avatars and signatures was a topic that had come up, and we feel needs to be addressed specifically. Avatars and signatures have had a long history that predates this forum: while some people use artwork made for them, many use images that they have found through other sources -- often these are either assets from other media, or from search engines.

As mentioned previously, search engines have been flooded with GenAI art. We don't believe it's fair to expect all members to know how to navigate this, and for those who struggle to, removing their avatar and signatures until they're able to find one that they like and which isn't GenAI seems like it would be frustrating. Many of the members who would struggle with this are those who don't have as strong technical skills, especially those who are in much older or younger age brackets.

Additionally, determining whether an avatar or signature was created with GenAI can be difficult at times. While there are cases where it's very clear, there have also been cases where it's been harder to tell, particularly due to the scale of avatars and thus the reduced detail. The effort required to make the judgement by the staff seems disproportionate compared to the harm done in allowing the avatar/signature to remain, and making a misjudgement and removing something that wasn't AI-generated may be frustrating for members to deal with.

Thus, we do not currently have plans on disallowing GenAI use for avatars and signatures. In an ideal world, the sources of all avatars and signatures would be properly credited and thus the use of GenAI for these wouldn't be possible. However, we don't have plans to change the culture of avatars and signatures, nor to enforce this as it would create a large amount of work for members and would be frustrating for all parties.

With that said, however...

Art credit and permissions
While crediting artists is good etiquette if you're using their art for your avatar or signature, it's largely left to an artist to decide on their terms of use: some artists are fine without credit being provided, others will only allow their art to be used with proper credit. Though we won't make a decision for all artists by requiring accreditation, we would like to make it clear that we will respect artists' requests if they wish for their art to be removed from an avatar, signature, etc. This applies to both artists that are members of the forum and those who aren't.

If you are using another artist's works, please be respectful of their terms and provide credit where it is due! If you're an artist who has seen your work being used on the forum without permission, please let us know!

 

[daringred_]

as someone who has commissioned and supported a lot of real artists, all i can really say is that this response is very disappointing to me. to expand on my stance:

i just think the response largely amounts to "we can't be bothered," or at least that's how it comes across to me. a lot of the instances of AI art in signatures/avatars on this site have been super obvious, and it's mostly users who bring it to the attention of staff rather than staff identifying it themselves, so i'm not entirely sure where i see there being too much extra work in that regard. (but, again, this is why we need more moderators or community liaisons.) nobody's asking for users to be warned/punished etc. if they use it, because chances are it's unintentional 99% of the time, especially as AI gets harder to spot, but just notifying them seems perfectly doable imo. some of the stuff in regards to threads, like concept art etc. just felt like whataboutism-ing your way into not monitoring its use. i feel like those examples are obviously different and should be reviewed/handled on a case by case basis.

 

[demoness]

at the very least you could require some documentation of the artists creative process during an event provided with the submission. i'm not talking identifying information; i mean like drafts. that isn't going to be practical for purely discussion focused threads, but you could eliminate some of the concern of unintentional and intentional theft where unfair advantage is relevant, the more stages in someone's process that can be verified. it could be argued that is too invasive for the artist, but i ain't arguing ethics, i'm just throwing noodles at the wall. im no artist or tech girl so i'm not gonna make a hardline stance, but it's not hard to get behind a balanced take like mistreils or even behind daringred, because it's not really a very big ask.

it could very well be true it may be impractical to go around confirming every case of ownership doubt, but to make a rule declaration that generative ai is disallowed on the forums isn't necessarily a requirement that the team has to surveillance every single post 24/7 unprompted as it isn't to say that it is necessarily wrong (how's that oxford butchery taste, writing amateurs) that necessarily right this minute today it is running rampant. but that isn't the argument: what it seems people want is more effort put into the checks and balances involved when something is suspect or when there is a noticeable competence inconsistency in a persons submission and sharing history, and if possible, that the staff could as a unit identify basic AI tells at the very least than be blindsided by a user ticking the boxes. that person doesn't even need to receive maximum sentencing the first time, but a proportionate response to the use of such, like in the minimizing of the reward compromise mistreil was talking about and/or if done in ignorance is corrected by the user publicly bc like no one is going to single out someone remorseful for the mistake. and it could also be an idea to allow said first time incidental violator to resubmit with the provision they won't win but could still be recognized. im not really an events type so idk how you guys run stuff, but nothing says you can't evolve to discourage ai use or like i said there are ways to

build events around more specific criteria or on something other than raw talent. and if staff is involved in any way in choosing winners or i guess "exalting" museum work, like i said i've never spent real time in those boards, then it sort of is just etiquette they have some responsibility to recognize when something is suspect. it is next to impossible for the layman using quality, paid programs but from what i read it doesn't sound the AI use that has been noticed is as difficult to parse? i ain't making objective declarations or anything, but i would like to suggest something this community never seems to tap into:

it's not a terrible idea to promote users documented expressing concern over something community positive to oversee that concern, or even ask a concerned group if a certain member has verifiable interest and ability in doing so. or you know, possibly take the advice of staff more qualified to understand generative AI and defer to maybe one who is an um. an artist, kinda moot now since the admin response seems to be correlated with major user exodus and terrifying the women and children, but there's always the future and giving people responsibilty maybe senior staff does not want anymore. i mean it happens. and to be clear i'm not talking about taking every psuedo-intellectual, lonely, handwringing narc up on their vapid caping into the team, because there are definitely pearl clutching squares adverse to even a grain of practically or catching up with the year they're in that fish for special consideration, but we don't necessarily need to make changes purely based on statistics. like i have staffed for sites with niche interest numbers and sometimes overstaffing doesn't have to be looked at as too many chiefs on one fire, it can also be a tool of scheduled efficiency

anyway that is just tangential thoughts the ai concern reminded me of.

now over this way where the goth birdie is: when it comes to avatars and signatures, i mean the straight edge solution is require how the avatar was acquired to be disclosed. and if you want to be even more sure, you could require a receipt of some sort submitted for approval before you can upload the thing. you could go even farther and disallow other intellectual property period. or reducto ad absurdum (to benefit) we find some rhythm between nuance and effort that daringred and even mistreil offered and avoid the angst and anyone else from making a hellish, waffling, contrarian pages long vapor just like i did, humans

soooo we *could* waste time with the idealism and the rhetorical posturing and the street level staff having to wrestle the pr and their actual informed opinion buried in the subtext and stuff, but to save time i translated my thoughts into the theatrics so i can boil it down to this:

i feel daringred 100% and i like mistreils informed practically

the queen demoness wants more of that and less chains of legalistic diatribes that don't have any critical thinking not just for this concern but for all the funny tiny mortals<3

 

[.: Meri :.]

I hope allowing AI images in avatars and signatures will be reconsidered. I think the increased normalization and use of AI-generated images (speaking of across the Internet in general, not specifically on TBT) is very harmful, given the ethical issues regarding art theft and the environment, and seeing them in avatars and signatures bothers me much more than when someone's signature is over the maximum allowed height.

 

[Mistreil]

Just dropping in real quick -- this isn't a proper/long response just yet; I've been waiting for other responses to come in first because I don't want to bias things or respond too soon. But my plan is to write something a bit more with my own personal thoughts/opinions on Monday after I get other work and stuff out of the way!

But I felt like I should quickly mention that the decision on "GenAI images are currently discouraged but not disallowed in other places (avatars, signatures, and other boards)" isn't a firm one, hence the 'currently'. It's something I'd be happy to bring up to the staff team for more discussion based on feedback received and people's perspectives on the matter!

In particular, it might be helpful to hear feedback regarding where you all would draw the line. i.e.:

• For AI images in posts themselves, what would be considered acceptable vs. not?
  (e.g. See the examples in the opening post - how would you want each of those to be handled, and what is the actual criteria you'd use for what should be considered actionable?)
• Thoughts on AI images in posts vs. avatars and signatures
  (e.g. feelings on "if AI images in one were restricted in one but not the other", whether it should be disallowed in both, etc.)

I'll keep my own thoughts on the matters until I have a chance to write my own likely stupidly long post because I don't want to influence things one way or another, but I do want to encourage people to provide their opinions and feedback!

 

[Aerith]

To the best of my knowledge, I have not used AI images and will continue to never use them on the forum, but despite being against AI, I understand why this is TBT's response. It's practical and moderator-conscientious.

However, I think there's value in taking a harder stance, because the staff is looking at moderating AI with nuance already. I have to wonder if what's holding them back might be worries of members' complaints. Members could criticize the staff for not being vigilant enough, that the members have to do all the work, or that the images aren't being removed in a timely manner. As long as members understand the rule is 'as reported', and these reports are dealt with within a reasonable time frame, there shouldn't be an issue.

I know there will be people who repeatedly use AI images, because they just don't understand what to look out for. Like Mistreil stated, it's easy to do when search engines are flooded with them. But, I feel that staff are empathetic and observant enough to realize who's clueless and who's intentionally being defiant. I don't think this should ever warrant a warning rather than a notification of removal, unless there's proof of intentional abuse.

What would be helpful is a pinned thread in the Bell Tree HQ titled: "Is this image AI?" There, more experienced members could help those who struggle to discern image authenticity, and if AI does get entirely banned in the future, this eases moderation burdens. Whether or not it's banned, I feel this also serves as a gesture of goodwill that the staff does not support the use of AI.

 

[oath2order]

I think a huge issue is exactly that, the determination of whether or not something is AI, and then having to deal with the back-and-forth of if something is AI or not. Generally, you can tell. But other times, it's harder.

 

[daringred_]

i don't think it's an unfair expectation to expect moderators to moderate. it seems they're in dire need of more moderators and/or to go the community liaison route anyway, and this is certainly something that could fall to the latter considering that it's mostly the community already alerting moderators to the use of ai art in profiles/signatures. i vehemently disagree that we should let it run rampant just because it's easier and "oh but it's just so hard to tell sometimes!" keywords here would be nuance and investigation. it's not as if we're asking for members to be punished/banned for using ai art -- as i already stated, chances are it's not intentional -- but if you're really not sure, you can just monitor the user or ask them where they sourced the image from. it isn't exactly rocket science. i've commissioned hundreds of artists across thousands of dollars at this point, and i've never once fallen victim to the ai slop. another option would be to mandate that users credit artists for profile/signature images (or at least link to the source) rather than simply encouraging it.

now if we're going back to this, nuance comes up again.

In particular, it might be helpful to hear feedback regarding where you all would draw the line. i.e.:

• For AI images in posts themselves, what would be considered acceptable vs. not?
  (e.g. See the examples in the opening post - how would you want each of those to be handled, and what is the actual criteria you'd use for what should be considered actionable?)
• Thoughts on AI images in posts vs. avatars and signatures
  (e.g. feelings on "if AI images in one were restricted in one but not the other", whether it should be disallowed in both, etc.)

it should be a case-by-case basis and whether or not there is malicious/trolling intent, for one thing. other members can easily notify them in the thread that, hey, this looks ai generated rather than conceptualized by an actual artist. posting something for discussion is less of an endorsement than having it represent you across the forum via your profile/signature, even if unintentionally. moderators could also add disclaimers to the top of these threads/posts noting that certain images are or may have been generated by ai. again, nobody is asking for others to be punished/banned over stuff like this. a discussion is different because users can discuss in the thread that it's ai and whether or not they would want to support the project based on that, educate the user who didn't realize and/or provide sources/art/images made by actual real people. like i said: nuance.

 

[Millysaurusrexjr]

I really hate AI generated imagery, I cannot say that enough. I have never used it nor do I ever intend to at any point in any of my art. I hate seeing AI generated images anywhere (which unfortunately is becoming an everyday occurrence, even off the internet). That said, a ban on AI generated images being used as profile pictures/signatures makes me feel uneasy. There have been witch hunts against artists, artists have been accused of using AI when they haven't. Even providing video proof is sometimes not enough for some accusers. Not to mention that AI continues to improve and it's getting more difficult to tell if something is AI or not.

Let's say AI pfps/signatures were banned on TBT -- If I were to draw my own pfp + signature, but didn't keep any sketches/WIP/timelapse videos/etc, and someone accused me of using AI, would I then not be allowed to use that pfp/signature?

Personally, I think it's enough to strongly discourage AI generated imagery ban on TBT. I'm not against an AI generated imagery ban, I am just worried about the possible scenarios that could arise.

p.s. If we do ban AI generated imagery from being used as PFPs/signatures, can the turnips from the 2023 april fools event at least be allowed? they're so ugly-cute