[Feature] Textual credit for images

[GreenReaper]

I’d like to suggest a feature mentioned in passing in a previous suggestion; image credits.

The Mucklet terms specify that credit should be given for other people’s content:

Respecting others’ rights
You may not provide content that infringes on someone elses [sic.] intellectual property. If you use an image, or add content based on someone elses work, make sure you have the rights to do so. And give credits where credit is due.

Currently I do this to the extent possible, and I have seen a few characters doing something similar, but overall there is a tendency to say “credit to the [unnamed] artist”, possibly with an offer to take it down. This isn’t fair to the artist or photographer concerned and not in line with the terms; and doesn’t clarify when a work is made by or for the character owner.

One reason for this may be that there is no default way to supply this credit. I suggest text input fields on image submission or perhaps as part of the room/character profile editor (though this would reinforce the already non-ideal “one image per profile” limit) for creator name, work name and license, with associated URL fields, all optional.

There might also be checkboxes for artist or commissioner. This way you could say it’s a commission and identify the artist, but not the work if it wasn’t public or didn’t want to out yourself as the player behind the character - or vice versa if the artist desired anonymity.

This would ideally replace what I currently use for credit, along the lines of:

--🖼️ [David Revoy](https://en.wikipedia.org/wiki/David_Revoy): _[Tree house](https://commons.wikimedia.org/wiki/File:2012-08-24_Tree-house_by-David-Revoy.jpg)_. [CC BY](https://creativecommons.org/licenses/by/4.0/deed.en).--

This renders as: David Revoy: Tree house. CC BY.

Or in Mucklet:

Because I put this at the bottom of the description I use an icon like or to indicate what it refers to (or when credit is due for text). This may not be necessary if displayed directly beneath the image. (If space is low I just use “BY” for the license, or “BY-NC-SA”, etc.)

I could also see it as an info line beneath the expanded view on-click, although there should be some indication of credit on the default view too, if not a line of text then an icon/mask.

[Kathrynn]

I’m going to open the can of worms and ask the following: How would I be able to tell if an AI image is using art properly and with consent?

I’m all for adding a by-line for your art, but the wording of current rules makes most art unusable.

[GreenReaper]

A lot of people do use AI-generated artwork. Personally, I try to find an existing image first, ideally one under a free license. I see many others doing at least the first of these but often without attribution; this is an attempt to address crediting of those images in particular.

Depending on where you are, AI art either does not count as a copy (further details), or does (if you generate the exact lyrics); and so doesn’t require consent/credit as it is non-infringing (notwithstanding trademark issues), or does to what is likely an impossible degree beyond mentioning a model which provides a list of sources (compare your lifetime browser history).

If you do want to credit AI art, I’d suggest that the model and version used is a good start.

[Kathrynn]

@GreenReaper

or add content based on someone elses work,

It’s this part that is the issue. Legal ruling aside, AI is based on someone else’s work. So … that’s a problem.

[GreenReaper]

Yes, but (LoRAs and other exceptions aside) only in the sense that an artist draws like other artists they have seen over the years. AI models don’t keep the original training material. They record something akin to “how do you see a lion that isn’t actually there in a cloud?”, based on repeatedly removing randomly added noise from a zillion images of lions.

I honestly think it’s more of a problem in this case to reproduce exactly the same image that a single artist created without crediting them or seeking their permission. If nothing else, that is specifically based on some-one’s work, not everyone’s.

[Kathrynn]

I’m not disagreeing with anything you said. That is how I feel as well. However, that doesn’t change that the way the rule is written now, this isn’t clear.

We want to target art used without consent, or at least give credit, where it’s due, so target that. Vague rules lead to vague enforcement.

[Nirofur]

I think that the only correct option is to install a checkbox along with the credit line.

[( ) Check if using AI-generated imagery]

As to the question of ‘consent’? It’s better to assume AI art does not have any artist’s actual consent. All AI image models are trained on data that was never consented to being used for the purpose.

[GreenReaper]

As an addendum: credit links can be very long, a problem for builds, abouts or character descriptions which hit the ~4000 character limit.

BUT! Wikimedia has a URL shortener. And Wikisource has Creative Commons licenses. So you can instead do something like this:

--🖼️ [David Revoy](https://w.wiki/RdnQ): _[Tree house](https://w.wiki/Rdnj)_. [CC BY](https://w.wiki/Rdpf).--

If linking to this forum, you can also eliminate the title text, e.g. https://forum.wolfery.com/t/3502 redirects here.

(Maybe one day we’ll be able to take the https: off)

[Pecus]

I don’t really like the idea of intrusively tagging everything with scholarly resource style citations for all your images. It’s distracting at best, and onerous on content creators at worst. If you want to see less content made, making that mandatory and standardizing it is definitely the route to accomplish it.

[GreenReaper]

If you created the image, there’s no requirement to credit it. It’s only if you use someone else’s work - i.e. not just “your image” - which deserves credit.

With respect to AI in particular: if that was intended to be covered by this rule, relevant information is often included with an uploaded image’s metadata, at least enough to identify the source, so the work could be done by the platform. Textual AI content is trickier (and perhaps more likely to be edited).

As for obtrusiveness, I imagine some small faded icon in the corner that displays text on hover and expands on click might work. Right now the best you can do is put it in an empty-titled section, e.g.

[[]]{...}

which displays as just ▶ (without a background). Personally I think this is a little too easy to miss for other people's work, but it does satisfy the terms.

[WakuWaku]

I appreciate your dedication to crediting images @GreenReaper. I’m neutral on the use of AI here legally and don’t want this to devolve into an AI discussion. I’d like to see building standards formalized to include image accreditation. I assume it’s something the builders look for upon review and having it added by the user just puts one less worry on the builder’s plate.

On implementation it should prompt on upload and display underneath a clicked image like the forums here.

[jmac32here]

@Nirofur

All AI image models are trained on data that was never consented to being used for the purpose.

This.

But worse yet, they are trained to create their “own” styles using the styles of EVERYTHING that’s ever been posted since at least 2006, when Google got analytics installed by default into nearly every piece of web server software.

At the same time, it’s called “trained” because the those posts are supposed to be used “as inspiration” for the ai models and not stored once processed.

[jmac32here]

Of course, recent court rulings are allowing AI to be trained in this manner with “any publicly available sources” (anything online) and when coupled with previous copyright rulings that stated “enough difference” (5%) meant it wasn’t in violation of IP laws.

The newer rulings do have one caveat, AI models and their parent companies cannot retain source materials beyond the training and cannot attempt to claim any IP rights of generated content.