Topic: Is tracing an Ai-generated content allowed on e621?

Posted under General

There's a "new" artist just posted on IB (CW: cub on IB)
I uploaded one of their pics as this, but mods unapproved and went:

spe said:

finnick621 said:
AI unapproved huh, that's weird cuz I checked them on Hive Moderation and their art was said to be legit.
Man how do mods even tell if they are Ai-generated

With our eyes. And a collection of data about identifiable AI trends.

Those "AI detectors" are extremely unreliable, btw. Very easy to bypass their detection just by generating less popular styles, or simply drawing over the original to make it look human-made (as in this case) - and perhaps worse, they very often give false positives on completely legitimate artwork, particularly from the kinds of highly skilled artists that most popular AI datasets are heavily trained on in the first place. Don’t bother trusting those for a second.

I believe staff have vast experience with distinguishability in AI-generated content. Far more experienced than I am at telling that the artist had traced some Ai content.
1) If a post is traced from Ai-generated content, Why does e621 ban that? Yeah I do know that e621 aren't allowed tracing another artist's work, but technically tracing an ownerless Ai content shouldn't bother any artists right? Like this post, I can't even tell who's the original in this case, assuming this is traced an Ai content.

2) If a post isn't traced but you claim it is, how can the artist prove otherwise? I am asking this one because the artist did tell they didn't use AI, of course this is a statement without their proving. I was just ensuring that every possible situation would have a reasonable explanation.

Updated

finnick621 said:
1) If a post is traced from Ai-generated content, Why does e621 ban that? Yeah I do know that e621 aren't allowed tracing another artist's work, but technically tracing an ownerless Ai content shouldn't bother any artists right? Like this post, I can't even tell who's the original in this case, assuming this is traced an Ai content.

Well with AI stealing from multiple artists it's even worse right? You are tracing basically at multiple artists at the same time lol. Also it could open the precedent of someone using AI to "slightly change" someone else art and then tracing it.

finnick621 said:
1) If a post is traced from Ai-generated content, Why does e621 ban that? Yeah I do know that e621 aren't allowed tracing another artist's work, but technically tracing an ownerless Ai content shouldn't bother any artists right? Like this post, I can't even tell who's the original in this case, assuming this is traced an Ai content.

Yeah, this AI ban has little to do with concerns of plagiarism at this point. We are not letting AI-generated artwork tracing or paintovers back at all, see topic #47144.

We did previously allow full AI paintovers to be posted early in the year, but it has proven to be more of a hassle trying to sort through artworks that were completely AI-generated, paintovers, partial paintovers, and actual genuine artwork. This is also on top of the many times people also try to contest their deletions and complain about it.

Since our Uploading Guidelines already says "No AI generated, or AI assisted artwork" and the site's main purpose is catered towards actual human-drawn artwork, the admins just decided to ban the whole lot of it.
The only exceptions allowed for AI on artworks now are on backgrounds and audio, and as AI-generated image references that are not directly placed over the canvas.

If an artist still wants to post AI-generated or assisted artworks, they are highly advised to post it on https://e6ai.net/ instead.

2) If a post isn't traced but you claim it is, how can the artist prove otherwise? I am asking this one because the artist did tell they didn't use AI, of course this is a statement without their proving. I was just ensuring that every possible situation would have a reasonable explanation.

The decision to blacklist/nuke an artist for using AI isn't done lightly. There must be a lot of evidence to suggest that they do use AI versus evidence that says they don't.

The fastest way to disprove this is for the artist to record a video of their work process, as well as proving the raw project files.
I believe some art programs also have built-in time-lapse features that an artist could use to record their work process.

Updated

thegreatwolfgang said:
Yeah, this AI ban has little to do with concerns of plagiarism at this point. We are not letting AI-generated artwork tracing or paintovers back at all, see topic #47144.

We did previously allow full AI paintovers to be posted early in the year, but it has proven to be more of a hassle trying to sort through artworks that were completely AI-generated, paintovers, partial paintovers, and actual genuine artwork. This is also on top of the many times people also try to contest their deletions and complain about it.

Since our Uploading Guidelines already says "No AI generated, or AI assisted artwork" and the site's main purpose is catered towards actual human-drawn artwork, the admins just decided to ban the whole lot of it.
The only exceptions allowed for AI on artworks now are on backgrounds and audio.

If an artist still wants to post AI-generated or assisted artworks, they are highly advised to post it on https://e6ai.net/ instead.

The decision to blacklist/nuke an artist for using AI isn't done lightly. There must be a lot of evidence to suggest that they do use AI versus evidence that says they don't.

The fastest way to disprove this is for the artist to record a video of their work process, as well as proving the raw project files.
I believe some art programs also have built-in time-lapse features that an artist could use to record their work process.

Wait. AI assisted backgrounds are allowed?
Shit. I just flagged 2 images for using AI assisted backgrounds. >.<

fuzzy_kobold said:
Wait. AI assisted backgrounds are allowed?
Shit. I just flagged 2 images for using AI assisted backgrounds. >.<

The specific quote to the exception is:

  • "Exceptions are currently for backgrounds (treated like using a photo as a background, quality rules apply); for artwork that references, but does not directly use, AI generated content; and for audio in video posts such as WebM."

That's why we have the ai_assisted tag, alongside its subtags ai_generated_audio, ai_generated_background, and ai_generated_reference.

If you have flagged the posts, there is nothing you can do now. A mod will probably go through them and inform you that they do not violate any rules.

notknow said:

thegreatwolfgang said:

Okok I now know the regulation about tracing AI content do exist.

But is visual painting details sufficient to support the charge in this case. The existence of a regulation and whether a case violates it are two distinct and independent things. I was unable to identify any intuitive artifacts of AI-generation in their artworks.

###
What if, I'm not using CSP or SAI to directly trace AI content on a layer above, but I manually trace AI content by referencing them on a split-screen or a second monitor. And I write a description inform you that I am tracing. Just like transcribing, is it legit?

Updated

Aacafah

Moderator

We can tell the difference between eyeballing an AI reference & tracing an AI reference the same way we can tell the difference between doing the same with a non-AI reference. We do have additional indicators we look for, but considering the potential for malicious users to specifically avoid them after learning what they are, we consider it ill-advised to publicly disclose them.

aacafah said:
We can tell the difference between eyeballing an AI reference & tracing an AI reference the same way we can tell the difference between doing the same with a non-AI reference. We do have additional indicators we look for, but considering the potential for malicious users to specifically avoid them after learning what they are, we consider it ill-advised to publicly disclose them.

Huh? So you meant you, the staff, can tell the difference among:
1) eyeballing an AI reference
2) tracing an AI reference on layer
3) eyeballing an non-AI reference
4) tracing an non-AI reference on layer
I understand the principle of using discretion to maintain a clear distinction. You can't tell a knockoff what makes it a knockoff, that would help it improve. But you didn't mention which of them are allowed on e621, it can't possibly be all of them right?

finnick621 said:
But is visual painting details sufficient to support the charge in this case. ... I was unable to identify any intuitive artifacts of AI-generation in their artworks.

As @Aacafah and @spe already said previously, there probably is sufficient evidence to suggest that the works were AI-generated, or at the very least, AI-assisted.
Whatever that evidence might be is based on "a collection of data about identifiable AI trends" which they are not going to disclose publicly.

If you still don't believe it is justified, you may refer to topic #23578 on the proper procedure to contest a deletion.
However, considering an admin has already clarified the issue with you, I don't think you would get a different answer.

###
What if, I'm not using CSP or SAI to directly trace AI content on a layer above, but I manually trace AI content by referencing them on a split-screen or a second monitor. And I write a description inform you that I am tracing. Just like transcribing, is it legit?

It is important to note here that "tracing" means you are copying it line-by-line without making any changes, so it will be virtually identical to the AI-generated image.
This is not the same as you "eyeballing" or "sketching" it based on an image on a second screen.

If you had made an artwork and verbally mentioned that you "traced" it from an AI-generated image, it will probably get deleted regardless if that statement was true or not.
You would also get smacked with a Posting Abuse violation due to knowingly uploading content that goes against the Uploading Guidelines.

If you had actually eyeballed it and drew the artwork while referencing an AI-generated image on a second screen, it is allowed as an exception "for artwork that references, but does not directly use, AI generated content."
This falls under ai_generated_reference, but you don't need to tag/disclose it unless you had inserted the AI-generated image reference itself within the post as well.

I despise artificial intelligence and theft from artists.

I have personally commissioned artwork and know it was drawn by a human artist without AI. And then when the artist posted it, staff took it down as AI generated.

It's hard for me to trust their judgment on the subject.

Aacafah

Moderator

grizzly_yote said:
I despise artificial intelligence and theft from artists.

I have personally commissioned artwork and know it was drawn by a human artist without AI. And then when the artist posted it, staff took it down as AI generated.

It's hard for me to trust their judgment on the subject.

You're welcome to contest this through the aforementioned methods. We aren't claiming to be the arbiters of truth here, but we don't make these determinations flippantly; we'd be happy to reevaluate the case if you or the artist can provide us with additional evidence, but as you can imagine, we have to rely on indicators that can appear in legitimate works. In fact, if you're confident we made the wrong call, proving that to us would help us make better decisions in the future.

thegreatwolfgang said:

I do appreciate the mods' use of discretion to reduce their workload, that's reasonable since we users basically don't pay them.
Our users’ trust in the feasibility of this discretion is based, I believe, on the fact that the visual distinction between AI-generated images and non-AI-generated ones is relatively clear to the majority, currently. But as AI develops, it certainly has become harder for us users to tell the difference between these images. Would this dilemma also apply to mods given that they may likewise find it harder? If this was the case, then doubts about this discretion would only get more common as my case and

grizzly_yote said:
It's hard for me to trust their judgment on the subject.

aacafah said:
You're welcome to contest this through the aforementioned methods. We aren't claiming to be the arbiters of truth here, but we don't make these determinations flippantly; we'd be happy to reevaluate the case if you or the artist can provide us with additional evidence, but as you can imagine, we have to rely on indicators that can appear in legitimate works. In fact, if you're confident we made the wrong call, proving that to us would help us make better decisions in the future.

Thank you, I did try that. The staff in question was unmoved.

grizzly_yote said:
Thank you, I did try that. The staff in question was unmoved.

Honestly, the artist probably needs to be the one to make the appeal rather than you. A commissioner is not going to have access to the kind of evidence we need to verify their legitimacy.

Their artwork was all removed due to containing a lot of the AI indicators that we look for - and while it is possible that their art just looks like that, it’s very unlikely unless they can provide some pretty strong proof of their legitimacy. Our doors are always open, however, if they want to try to make their case.

On a side note, I understand how hard it would be to accept that an artist that you paid to draw something for you was just tracing AI gens. There was a case a while ago where twinkle-sez admitted to using AI in some of their commissions, but offered to redraw their commissions by hand after coming out about it, which was cool. But this kind of thing really does happen often and you can surely understand why we’re not going to accept some artwork that we think is AI assisted just because you paid for it and don’t believe that you were paying for AI assisted art.