Skip to main content

Indie game storeFree gamesFun gamesHorror games
Game developmentAssetsComics
SalesBundles
Jobs
TagsGame Engines

Generative AI Disclosure tagging

A topic by leafo created 1 day ago Views: 4,824 Replies: 24
Viewing posts 1 to 15
Admin (6 edits) (+21)

We’ve deployed a new field to project edit pages called “Generative AI disclosure.” This field asks if your project contains the results of generative AI. If you select yes, then we also ask what kinds of generative AI you utilize, including Graphics, Sound, Text & Dialog, and Code.

Screenshot of AI Generated disclosure fields

If you select “yes,” then your page will automatically receive the AI Generated tag.

Additionally, the following sub-tags are used for the content types:

For projects that select “no,” they will receive the tag No AI.

Note: It’s not necessary to use these “automatic” tags manually when classifying your page. We may rename or modify them to fit future classification needs, so please correctly fill out the disclosure form instead of manually tagging.

Required for Asset Creators

This field is now required for all asset creators on itch.io. If you have a public asset page on itch.io and you view your dashboard, you will now see a blocking dialog instructing you to classify your pages.

Screenshot...
Screenshot of bulk tagging tool dialog

Because we understand that some asset creators can have a large quantity of pages, we’ve included a bulk tagging tool to simplify the process of tagging assets.

Assets comprised of generative AI (even if modified afterwards) that are not tagged will no longer be eligible for indexing on our browse pages. You can review our quality guidelines here: https://itch.io/docs/creators/quality-guidelines#ai-disclosure

We’ll have a grace period for people to update their pages, then we’ll likely use user reports to handle pages that have not been addressed.

(+2)

This is very helpful, finally I can filter out the AI stuff. Thanks a lot :)

However, I don't see the "No AI Tag" in my project infobox after setting it up. Am I missing something?

(+1)

From a design philosophy and as a user, I would not want to see a "no-ai" metainfo in that box.

You also do not want to see a "not made with unity", "not made with rpg maker", "no touchscreen input", and so on.

Leafo called it tag, and it is in the suggested tag box, and it shows as an url modifier with tag, but it seems to be metainfo. Like "made-with-godot".

I see, thanks. Yeah the term tag confused me.
It would still be nice to be able to view meta-information on the project page, since not everyone gets there via the search function. But that would be another topic.

(+2)

Me too. And that tag was a regular tag before as well, as far as I know. It was bound to be, what with Itch's free range tagging system.

So yeah, I expect the info box to show the meta info about tool usage in the future. There seems to be much development in that regard. Itch now autodetects certain engines as well.

I just hope they will not show the no-ai as info in all the games. Maybe for assets it is appropriate. For regular games it would be distracting. It should show the usage of AI, but not the non usage.

(+4)

That there is a no-ai tag is a very good idea.

Some thoughts and questions about this new feature.

Will the sub tags have negated tags as well? I guess, most users would want to use this functionality to not see AI projects. So, like browsing for no-ai-text + no-ai-images + no-ai-sounds to not exclude all the games that used llm to generate a basis for some of their code.

The AI debate mainly focuses on anything that is called art. It is most important to disclose it for assets, since they in turn would make the projects that use them be AI using projects. But funnily enough, code does not seem to be considered much in those debates. Coders also typically do not call themselves artists. But AI usage seems to be rampant in modern software development.

Also there is certain filtering and editing options in the big graphic editing softwares that use AI tech. So from a purist's view, usage of Adobe Photoshop would constitute usage of generative AI content tool. They even have functions that have generative in their title, like generative fill that works with a prompt.

So maybe some clarification would be in order in the faq. There are people that take words literally, and if you cite music and they used ai-voices, they might think to be exempt - maybe they even are, since an AI voice does not use a prompt. And if you cite the major llm products and they "only used Photoshop" they might also not grasp that using a prompt to do something is generative AI usage, no matter how it is called in your editing tool. And there are functions that do not use a prompt, yet generate content - generative expand at least has generative in its name.

In other words, make the wording as non ambigous as you can, so even people with little or no knowledge English that see an AI translated version will understand . You might avoid some people asking support, if their thingy is considered AI or not.

Oh, and will we see the AI meta info in the information box? Currently there is no information. Only assets/games, that have this as an actual tag, have it displayed.

Admin (4 edits) (+9)

Negative filtering is coming soon.

The no-ai was created for convenience, as it is not a strict negative of ai-generated. The no-ai tag is only applied to pages where the creator explicitly marked their page as not containing generative AI. Pages that haven’t completed the tagging will not receive this tag. Think of it as a way for a creator to announce their page is AI free.

Developers browsing for assets will love this feature! Even more, once the information box will show the AI usage meta info.

And since that tagging is mandatory for assets, after that grace period, they will either be tagged yes/no or be deindexed anyway because they did not give the information. So actually the no-ai should evolve to be a strict negative for assets.

But I fear that being "AI free" on code will be a grey area. Positivly claiming you did not use those llm helpers should be ok-ish. People usually know what tools they used. Mostly. I am not so sure about a thing like Photoshop. But even using a translator service on the net might constitute llm AI usage.

To me anything you use a "prompt" to create something is the "AI" we are talking about. That is why I am unsure about AI voices or translations. And even code snippets might be debateable. Or using a llm to debug code, as mentioned below.   

Please make AI assets opt-in rather than opt-out when negative filtering comes!

(+3)

I had voiced despair at how flooded with AI the assets section had become, and even stopped publishing on itch altogether. This is very good news, a good start, and I'll definitely come back to itch if negative filtering does a good job of filtering out the slop.

I dont mean to sound ungrateful here but why not ban AI entirely? No one wants it, and it's just a shitty grift that actively harms the indies that your platform specifically caters towards. Itch has always been a haven for devs and artists, why even allow assets that have had no significant human editing? Moderation seems to be stretched thin, and that also seems like a faster way to get the issue done with

(+2)

Will there be any punishment if someone lies? I think that a good portion of people know that people will filter No AI so they will just lie when setting the tags.

Admin(+3)

Our same quality guidelines apply, if you abuse the tagging feature to lie then you will no longer be eligible for indexing on our discovery pages.

https://itch.io/docs/creators/quality-guidelines#do-not-use-unrelated-tags-or-classifications-to-promote-your-game

One of my titles has been flagged even though it does not use AI. How do I appeal this?

I noticed that this applies to code as well. So, if I use ChatGPT to debug and help me with code, do I also need to include a tag indicating it was made with AI? If that’s the case, I believe there will be a lot of games with the AI tag if developers are honest about it. If the goal is to filter out bad assets, I don’t think this will help much, considering the majority of games would end up with the tag. However, if it’s just to inform players, that’s fine. I’m just asking to stay informed.

(+1)

If you have a public asset page on itch.io and you view your dashboard, you will now see a blocking dialog instructing you to classify your pages.

When I go to my dashboard, I don’t receive any dialog. I can edit the AI settings on my individual project pages, but I have quite a few projects and would much rather update them in bulk. Is there a way I can force the bulk dialog to show?

(+2)

It's because it only appears for assets, not for other products. But, I seconded this because I'd love for the bulk tagging tool to be available for all users.

I still have the AI option on my projects that aren’t assets (none of mine are assets), so I can edit the setting per-product. I suppose I’ll do that eventually, but it will take a little while.

If I'm not mistaken, the reporting is only mandatory for asset packs at the moment, right? Will games and other projects have mandatory AI disclosure at some point in the future?

(1 edit)

Both of my games show the option to display if it was made with IA or not and I saw other games in the "Made with AI (code)" in the post link. I think that is for everything.

I think enforced disclosure is the most sustainable way of dealing with AI content long term. I completely understand people wanting to make informed decisions about where the content they consume comes from, and I wouldn't intentionally use AI assets in my own work.

Banning it completely could lead to human creations being unfairly penalized for perceived resemblance to AI (as has happened in academic settings), and removes the possibility of use for criticism and review—or even honest mistakes, as people not educated on the details of AI may not realize some of the tools they use could be considered AI.

Plus, there are legitimately creative human works, including on Itch, that have used AI assets. The early builds of The Roottrees Are Dead, for instance, used AI art. While I think the AI art was the weakest part of the game, the game was fully written, coded, and performed by humans, and the version now in development with fully human-created art may not have existed without those early builds.

So I think this is a healthy choice, and I'd like to see more clarity on what "generative AI" is considered to be for the purposes of tagging, as well as what the appeal process would look like if work is incorrectly reported as untagged genAI.

human creations being unfairly penalized for perceived resemblance to AI

That they trained the models by real artworks does mean, that there are people out there that can and do create artworks that look like that...

I worry more about the code AI disclosure. In software development, AI tools were embraced more than they were shunned like in visual art. At least this is my impression. It is quite normal to import code snippets from elsewhere and modify them. Creating that snippet with a llm, instead of spending hours on dedicated message boards and documentations sounds like it would be business as usual and not even be perceived as the evil AI usage everyone is talking about.

To ban (generative) AI, it would have to be illegal to begin with. Banning it because some people do not like the concept or call it unfair or whatever would open the door to banning things for all kinds of reasons. When digital art creation was beginning to emerge, there also was a step that created unfair advantages. Imagine a place where art would be banned, because it was not done with pencil on paper, but with pixels and a program that would automatically fill in colors in a place.

Anyway, ther reason stated in the quality guidelines currently reads like this: legal ambiguity around rights associated with Generative AI content

If you sell an asset, that is a thing to be used in a thing that is to be sold yet again. If the legal situation has ambiguities and you do not know, if the asset you bought gives you full rights in the future for the content in said assets, that is a problem. Worst case, you try to make merchandise from your successful game and the genAI content you bought resembles an actual IP, since it was trained from data that containted that IP.

So I fully understand and welcome disclosure for products that are to be used in other products.

As for disclosure of AI in games, I would very much like to know what Steam does with that info or plans to do. They do collect the info since summer I believe. But I have yet to see a game description on Steam that told me, that a game uses genAI content. Maybe I just missed those. There also does not seem to be a filter option to not show games that have genAI.

What would be of course not ok, is to brag about hand made pixel art and whatnot, while the game is genAI. That is like selling organic food that ain't.

(+1)

How long is the grace period?

(1 edit)

"Assets comprised of generative AI (even if modified afterwards) that are not tagged will no longer be eligible for indexing..."

...How will untagged assets be identified, especially if modified after?  I didn't think we had tools that could accurately identify all AI-generated content. (We definitely can't identify AI-generated text, but that is less likely to be included in asset packs. )

I can definitely see hostile reporting of assets being an upcoming issue.  What will be used to decide if a project does or does not contain AI-generated content?

Does this apply to GitHub Copilot as well? I don’t think code should be part of this policy. Nearly every software developer uses AI tools now - whether for code completion, debugging, or generating boilerplates. It’s simply an efficiency tool.

With “AI Generated Code”, it might be a very difficult thing to know.

For example, if I use a library, and the authors of that library used AI to pad out unit tests - none of that code ends up directly in my project, but you could definitely argue that GenAI has benefited that project, and so mine.

And whilst I’ll probably know if a big library I’m using does that, what if it’s a random dependency five levels deep?