Topic: Young ferals

Posted under General

Adusak90

Member

As a feral enjoyer I'm gonna go straight to the point here: there's no reason to ban young ferals and especially ones that just look young.

I know the reason is that some of the pictures could be somehow based on an illegal pictures (which let's be honest, were probably filtered out by the OWNERS OF THE AI BASE TO NOT GO TO JAIL, but that's besides the point), but how in the hell feral four-legged creatures could be based off human pictures? There's no possibility for something like that.

I'm so sick of cute eevee pictures like this one https://e6ai.net/posts/142308 getting sent into the shadow realm because the young human policy was for some ungodly unrelated reason stretched onto ferals. Anthro I could understand, two legged creatures could have their generation include humans, okay, but how is a feral eevee supposed to do that? Especially in that normal dog-like standing position?

Good thing I automatically download eevee/mew/fennekin/[insert any first evolution four legged dog/cat/fox pokemon] pictures because the amount of times a picture would've been lost forever for me otherwise is worringly high.

kalethOrebiter

Privileged

To answer I need to clarify a few misconceptions.

1. Furry artwork is anthropomorphizing animals; in short, it's giving human qualities to animals.
- For content to be considered non-beastiality, ferals must have humanizing trait (often facial expressions). If they don't have human traits, that's bestiality/zoophilia.

2. For ferals, anthros, humans or anything in between: young is young, no exception.

3. Children and animals (not ferals; animals) are assumed to be unable give consent.

---

adusak90 said:
there's no reason to ban young ferals and especially ones that just look young.

False: see reason #2 and #3

adusak90 said:
the reason is that some of the pictures could be somehow based on an illegal pictures

False: False: see reason #2 and #3

adusak90 said:
how in the hell feral four-legged creatures could be based off human pictures? There's no possibility for something like that.

True, but that's irrelevant.

adusak90 said:
I'm so sick of cute eevee pictures like this one https://e6ai.net/posts/142308 getting sent into the shadow realm

That's understandable, but those picture can be host on other websites; e6ai staff and admins decided is just not the place for that content.

adusak90 said:
Anthro I could understand, two legged creatures could have their generation include humans, okay, but how is a feral eevee supposed to do that? Especially in that normal dog-like standing position?

True, but that's irrelevant. The decision is not based on hypothetical use of illegal material; it's based on the things that are contained in the post.

Hammapajamma

Member

What I don't get about this stuff is the inconsistency. What makes this Evee a particularly young character, when this one? apparently isn't? They're both Eevee drawn like an Eevee, with the same cutesy eyes-to-body ratio and compact build and everything. It just seems arbitrary at times.

How are you supposed to know what you're allowed to upload in that sort of environment?

Updated

kalethOrebiter

Privileged

hammapajamma said:
What makes this Evee a particularly young character, when this one? apparently isn't?

That just mean you found 2 edge case images.

hammapajamma said:

It just seems arbitrary at times

What do you expect? Janitors are humans and each artwork is different: there is bound to be variety from janitor to janitor, from artwork to artwork, from minute to minute.

The staff is well aware of this, and this is why the members are allowed to appeal deletion, and this is why no approval are final.

hammapajamma said:
How are you supposed to know what you're allowed to upload in that sort of environment?

Make sure you generate things that doesn't flirt with the limit defined by the rules.

Hammapajamma

Member

kalethorebiter said:
That just mean you found 2 edge case images.

Not... really? There's a whole bunch that are similar to that one because it's just the default Eevee look, so it tends to get generated like that. Same goes for Pokémon like Vulpix and certain other cub-looking stage 1 pokémon.

kalethOrebiter

Privileged

hammapajamma said:
cub-looking stage 1 pokémon


You said it yourself: cub looking.

If you don't think that "cub looking" is where the edge cases will be, I really don't know what to tell you.

Updated

sergalbutt

Janitor

hammapajamma said:
What I don't get about this stuff is the inconsistency.

This was spoken about recently on the e6ai channel on Furry Diffusion.
I'll link to the start of the conversation if you'd like to read through it yourself, but if you don't want to, I'll also summarise the key pointers that were made (mostly by ceti).

Different staff members can enforce the same rules differently. When it comes to "young" posts, we have had constant and ongoing discussions about them, and while we do have some general agreement of what does and doesn't constitute "young", there's still a bit of "vibes-based" judgement when it comes to whoever staff member is left to either approve or delete a post.

Now, here's an important note: Uploaders are allowed to contest those deletions, just as they can if it was any other post deleted for whichever other reason. It's in the site rules, but I'll put it here and highlight some specifics here too

Disputing Staff Actions

If you are ever faced with a staff action that you do not agree with – be it a post deletion, a warning, or a ban – the first thing that you should do is contact the staff member responsible for that action via a DMail.

You'll even find that most of us have that written on our user profiles, which are easily accessed via the deletion reason, or if you're on FD or the site's own Discord server, we're just as easily contacted there. We're not monsters, and we're more than happy to clarify our reasoning for decisions made, and if you still disagree, well contested deletions then get discussed by all the staff, and put to a vote. There have been countless posts that have been undeleted this way, including ones exactly as you and the OP were describing.

Is it a bit annoying? Sure, but if an uploader/director truly disagrees with said reasoning, then this is the best and fairest way to go about it.

I've said it countless times, but it's not as if we're actively looking to delete posts. We want people to upload their gens. There would be no e6ai if all we did was delete things, but there are rules and quality guidelines present that are expected to be followed, and it's us staff members who are there to make sure they are. We're individuals, not a hivemind, and that means that sometimes mistakes get made, but for when perceived mistakes DO happen, we also have rules and systems in place to help correct that.

Hammapajamma

Member

Thank you, knowing that the practical side of enforcing such a rule (likes when it comes to creatures that we know don't have to be as juvenile as they appear, which of course isn't strictly following TWYS but still a bit frustrating, because you can't just unlearn that knowledge) has at least been well-discussed and evaluated by staff is reassuring.

And I actually wasn't aware that deletions over content were open to dispute as much as the typical ones that are about anatomy flaws and other artifacts. Okay, that's nice. 👍

kalethorebiter said:

You said it yourself: cub looking.

If you don't think that "cub looking" is where the edge cases will be, I really don't know what to tell you.

I'm sorry, I thought you meant I was cherry-picking specific Eevee posts that happen to be edge cases (considering one of them was the OP's deleted upload), when in reality the whole species could be considered an edge case when on-model.

It's why I found it strange that quite a few have been approved in the past, even when accounting for different janitors having slightly differing views on what exactly is too young-looking.

kalethOrebiter

Privileged

hammapajamma said:
deletions over content were open to dispute as much as the typical ones that are about anatomy flaws and other artifacts. Okay, that's nice.

Yeah! staff is well aware that human decisions are flawed, and they will readily admit when they get something wrong.

Thing is, AI is pretty darn hard to tune for moderation, and still they make errors (just look at civitAI's auto tagger). Really, as of this date, the best system is still human curation/classification/janitors.

hammapajamma said:
I was cherry-picking specific Eevee posts that happen to be edge cases (considering one of them was the OP's deleted upload), when in reality the whole species could be considered an edge case when on-model.

Ah! I see! Indeed, I wasn't saying you cherry-picked them. In fact, I feel like it's an excellent example of how hard cub-looking pokémons/digimons/etc are to classify. I even agree with you: the 2 artwork are so similar that I couldn't tell you an objective reason for accepting one and not the other.

hammapajamma said:
It's why I found it strange that quite a few have been approved in the past

The thing is, it is not against the rules to post artwork containing eevees, riolus, (or any-subject-that-can-be-classify-as-babies), that are lore-accurate. What is against the rules is sexualizing "youngness" and young-looking characters.

It's a subtle distinction, but that what allows post like this one, and that one to stay on the website. On that post, there's no over emphasis on the fact the character is young.

(at least for now)

Kiyoka

Member

Not totally related but the whole cub thing is non-sensical regardless. It's an inconsistent rule between e621 and e6ai. Newer models past SD 2.0 didn't have the issue of being trained on illegal material due to using a subset of LAION-5B that didn't contain any pornographic material.

Another issue is token bleeding. Even if we assume it was trained on such content, making ANY sexual images should be forbidden considering those illegal images would bleed info into the generated image from pertinent tags even if specifically 'young' and similar tags weren't used.

At this point I feel like this rule is a relic and should be reconsidered. It made little sense in SD 1.5 that used unfiltered LAION-5B and the 2.0 version too. But for SDXL, SD 3.0, FLUX, etc. it makes absolutely no sense unless the owners of e6ai just don't like said type of art on a personal level which I repsect. It's their site to run as they please in the end.

Edit: Sorry for necro'ing I just wanted to see what the situation with cub art was on e6ai.

kalethOrebiter

Privileged

kiyoka said:
It's an inconsistent rule between e621 and e6ai. Newer models past SD 2.0 didn't have the issue of being trained on illegal material due to using a subset of LAION-5B that didn't contain any pornographic material.

I have no idea what you are talking about, janitors don't check which model genned an image for approval, that would be a huge time sink and impractical. Also, it is impossible to know which dataset a model is trained on by looking a genned image, even with the metadata.

As for inconsistency e621 and e6ai, both sites are administered independently, so it's not surprising there's inconsistencies between them. This site's admins decided to forbid sexualized cub/young content for legal reasons. The way it was created is irrelevant, the content is simply unallowed.

sergalbutt

Janitor

kalethorebiter said:
I have no idea what you are talking about, janitors don't check which model genned an image for approval, that would be a huge time sink and impractical. Also, it is impossible to know which dataset a model is trained on by looking a genned image, even with the metadata.

He's referring to the dataset that SD1.5 was built with and the original talking point here that young posts weren't allowed because of the very real thing that happened - the LAION-5B dataset that was essential the core building block back then was found to contain actual CSAM (and not an insignificant amount either)

As for inconsistency e621 and e6ai, both sites are administered independently, so it's not surprising there's inconsistencies between them. This site's admins decided to forbid sexualized cub/young content for legal reasons. The way it was created is irrelevant, the content is simply unallowed.

Actually, 621 and 6ai have one more layer above them - Bad Dragon (aka "management")
And it was management - not the admins - who made that decision, or so that's what I was told when I enquired about it a little while back.

kalethOrebiter

Privileged

sergalbutt said:
/snip

That's on me, I made some intellectual shortcuts and I stand corrected. Rightfully so.

So let me reformulate this:
It is impossible to know if CSAM was somehow use to gen an image, unless the janitors have at least all the following:

  • The image(s), and/or parts of images used in in images-to-images generation
  • All the metadata from all prompts for images that aren't raw gens
    • Metadata is often stripped or altered for a ton of reasons.
    • Metadata usually only have the prompt data from the latest generation.
  • Have knowledge or find information on what dataset have been use for:
    • the model,
    • the checkpoint,
    • all the loras,

So it is not strictly impossible, but it's not something you can expect janitors to do.

That is why I confidently said that janitors don't check which model genned an image for approval.

---

I conflated Bad Dragon/management and admins — also a shortcut I'm guilty off. I also stand corrected on the reasons young/cub porn art is forbidden. I couldn't debate at what level LAION-5B dataset is the reason for that rule (after all, it would not apply to e621, so there certainly other reasons)

Nevertheless, it doesn't change the fact that the content is forbidden, and there's no talk to allow that content at the moment, AFAIK.

Kiyoka

Member

kalethorebiter said:
it is impossible to know which dataset a model is trained on by looking a genned image, even with the metadata.

Metadata contains the model name so if you made a workflow on the site to extract the model string it should be quite easy to check, granted people upload the png that contains said metadata. Reading one word isn't a timesink, if it's not available and the content is questionable then they can remove it.

kalethorebiter said:
As for inconsistency e621 and e6ai, both sites are administered independently, so it's not surprising there's inconsistencies between them. This site's admins decided to forbid sexualized cub/young content for legal reasons. The way it was created is irrelevant, the content is simply unallowed.

Legally there is no issue unless they live in one of the very few places that have rules over artwork. And even then, it's really about where the server is located rather than where they live. But as I said, it's their site and they have every right to ban whatever content they want at their discretion. Let's just not pretend it's for some moral or legal issue that perpetuate bad connotations about AI use.

Kiyoka

Member

sergalbutt said:
Actually, 621 and 6ai have one more layer above them - Bad Dragon (aka "management")
And it was management - not the admins - who made that decision, or so that's what I was told when I enquired about it a little while back.

Thanks for clearing that out. It's sad that there is no debate on the issue as models have advanced past that. I guess they don't want to risk the bad rep that might come from people unfamiliar with the issue and be on the safe side. It makes sense from a business standpoint even if I am in disagreement with it.

One correction though, the content of CSAM in the dataset was quite insignificant. If I remember correctly there were on the lower end of a couple of thousand images in a dataset of 14 billion? That's less than 0.000002%. If that's not insignificant I don't know what is.

Aacafah

Member

kiyoka said:
Metadata contains the model name so if you made a workflow on the site to extract the model string it should be quite easy to check, granted people upload the png that contains said metadata. Reading one word isn't a timesink, if it's not available and the content is questionable then they can remove it.

Reading 1 word isn't a time sink if it's directly available & you're only checking that one word on that one post. Reading 1 word for ~125 uploads a day (at time of writing) when you have to send the post to an external tool to read the metadata (that also isn't support on WebM files)? That's a different story. Coding the functionality to do so directly into the site is also a time sink. Idk if you've noticed, but us e621 devs (there are no dedicated e6ai devs currently) are spread so thin we haven't updated either site in over a month (the version number at the bottom of every page is the date of the release; at time of writing, last was December 10th) when we typically do so weekly; do you think we have the time to add this unique functionality solely for our smaller sister site?

kiyoka said:
But as I said, it's their site and they have every right to ban whatever content they want at their discretion. Let's just not pretend it's for some moral or legal issue that perpetuate bad connotations about AI use.

Let's also not pretend that legal & moral issues are the only 2 issues at play.

Kiyoka

Member

j_mor said:
do you think we have the time to add this unique functionality solely for our smaller sister site?

I'm not a webdev and I won't pretend to know anything concerning the difficulty of implimenting such a feature and deploying it over existing ones or the issues that might cause. It was just an idea for a solution.
Could as well have people simply state the models they are using, and frankly, having either tags or a place to put the model used/loras in the description would be a nifty feature to have in case people like a certain style and the director wants to share it.(I know they can put it in the description but having a dedicated box for it might promote that type of behavior more).

As for what NotMeNotYou was saying, it's pretty much 1:1 what I also said, it's their business and they might want to avoid bad rep, that obviously was meant for their partners. And I perfectly understand that. I just said we shouldn't pretend that it's for legal or moral reasons as that fuels other kinds of diatribe online.

DRLa

Janitor

Had e621 been founded at the same time as e6ai, it would have most likely also forbidden young explicit content, because it's just not worth all the trouble.

It feels more like allowing young explicit is a legacy thing on e621 than the other way around on e6ai.

And I wouldn't be surprised if the rules for it got stricter in the future: defending this stuff is simply not anywhere close to mainstream acceptability, and never will (and to be honest I much prefer this content be hosted on dedicated sites I can avoid, because the legality of fictional non-human explicit young content is dubious at best where I live).