Danbooru

Make AI artist qualifiers necessary for AI artists

Posted under General

I'd be ok with this. However, we'd need to distinguish between artists that merely experimented with AI, and those that never made any illusions to being an actual artist. The latter group I'm ok with sticking a label on essentially saying "do not upload".

I like this a lot and think that someone uploading from the "artist" will automatically realize that it isn't human art with this. Most of it probably isnt intentionally uploaded so this should help.

+1 fully support this for any account that hasnt done any real art. Real artist who switch to AI shouldn't have the qualifer and are a separate can of worms.

I'm all for this. Personally, I think making it like banned artist is a wonderful idea. However, how exactly would we get "artists" implicated to this whenever someone accidentally uploads from a new one? Would that have it's own designated thread, or would the artist alias request thread be fine pulling double duty with this?

Maiden_in_Orange said:

I'm all for this. Personally, I think making it like banned artist is a wonderful idea. However, how exactly would we get "artists" implicated to this whenever someone accidentally uploads from a new one? Would that have it's own designated thread, or would the artist alias request thread be fine pulling double duty with this?

That thread already gets bumped to hell and back, and I think this is important enough to warrant its own topic anyways.

+1 for the idea, uploading any popular copyright nowadays is an AI-gen minefield and there's no reason not to lessen that risk.

+1 for the qualifier and +1 for the separate topic.

So many times I've stumbled upon nice looking posts with an existing artist tag, only to find out during uploading that this artist is an AI artist and realizing I've just wasted precious time. I also am heavily in favour of a seperate topic. Being branded as "AI artist" on Danbooru is quite a heavy accusation so a proper place for these discussions and implication requests is great.

I do want to propose that all implication requests made in this topic are required to come with some proper proof.
Think of outright mentions of AI/NovelAI/StableDiffusion in Twitter bios or comments (preferably with an achive.is link), pointing out artifacts in several images, links to discussions of the artist where they mention AI on other platforms like Reddit, as opposed to a single unfounded comment that boils down to "This feels like AI"

Reason being that this topic can then both function as a proper searchable papertrail for past and present discussions, and as a reference guide for people to learn what to look for when it comes to AI artifacts and red flags whenever they are in doubt.

The wiki page of an AI artist that has been processed can then be updated according to a template containing a link to the implication.

This artist has been marked as an AI artist.
Do not upload. See forum #283579

The searchability of the forum is vastly superior to artist wiki pages, so I'd prefer to keep the bulk of the information on the forum.

Some considerations:


Implications
Pros:
  • harder to abuse
  • more scrutiny on each request
  • ai-generated can be auto-tagged (with bots or via site integration)
cons:
  • bottleneck will be admins
  • cannot be easily used with extensions such as translate-pixiv-tags

Qualifier
Pros:
  • with translate-pixiv-tags the detection of AI prompter accounts is immediate and you don't even need to get to the upload page
  • no approval bottleneck
cons:
  • far less scrutiny, less eyes on potential vandalism and clueless users
  • artist creation bots (such as mine) cannot automate this, so we'd still have to fall back on BURs for mass renaming new artists created by bots

I don't mind taking care of the implications, but I'm afraid it would cause people to start creating tons of tags in advance without any upload just to discourage people from uploading from any obscure obvious AI prompt account they come across, and I don't know what volume of requests we'd get.
On the other hand, seeing what kind of pants-on-head retarded flags for "AI" we've gotten in the past, I'm really worried that just letting anyone "mark" artist tags as prompters would cause people to effectively poison certain new artists because they mistakenly thought it was AI, and there's naturally going to be far less eyes on new artist tags, especially if they're for the purpose of blacklisting. And that's not even getting into vandalism.

There should also be automatic rejection for attempts to rename/imply artists that existed before AI took off. Even if they started doing AI after drawing for real, it would mean their older, hand-drawn uploads would have tags indicating they were AI.

+1 for implications

nonamethanks said:

Implications
cons:
  • cannot be easily used with extensions such as translate-pixiv-tags

To mark banned artists, TPT depends on `is_banned` field from API.
This may be a niche case, but could it also be extended to return some kind of `is_ai` field - either calculated from the implications, or just straight up as a new field in the database?

hdk5 said:

+1 for implications

To mark banned artists, TPT depends on `is_banned` field from API.
This may be a niche case, but could it also be extended to return some kind of `is_ai` field - either calculated from the implications, or just straight up as a new field in the database?

You can use includes to get whether an artist implies another tag from the api: see this example. A hypotetical ai prompter tag would be returned in the same way. If TPT can be adapted to use this, then it would be one less reason against it.

nonamethanks said:

On the other hand, seeing what kind of pants-on-head retarded flags for "AI" we've gotten in the past, I'm really worried that just letting anyone "mark" artist tags as prompters would cause people to effectively poison certain new artists because they mistakenly thought it was AI, and there's naturally going to be far less eyes on new artist tags, especially if they're for the purpose of blacklisting. And that's not even getting into vandalism.

Big +1 to this. It needs some kind of oversight, users absolutely will mark actual artists as AI on some idiotic hunch. Getting marked as AI here could actually have a significant negative effect on their reputation, and I don't want that to happen. We've already seen lots of totally unjustified AI flags and witch-hunting and I don't want that to get any worse.

It's a million times better to have an AI artist accidentally accepted here as an actual artist than to have an actual artist get removed and bullied on twitter because of our mistake.

sabisabi said:

This sounds like a fine idea on paper, but I'm a little worried regarding artists like tarte_(hodarake) who used to draw and then started using AI. Will their previous hand-drawn works be automatically marked as AI as well?

I can vouch on that. I once commissioned an artist 10 years ago (amateur due to her having a day job) for free and what came out was decent, though the coloring was off due to her eye problem at the time. Some of her other works before and after that were equally decent as well. Unfortunately, that person is now using some AI in her latest works.

So yeah, we have to tread carefully when dealing with these artists.

1 2