Hey m@tes, as y'all know, this instance has been anti-corporate GenAI positive since it's creation and as such we've typically allowed such content to be posted freely. However in the last few weeks we've had a bunch of drama from GenAI haters who insist on coming into our comms and starting slap-fights. This caused us to vote on a new rule to have the mandate to clear out this constant friction. This worked to an extent, but I think we can help foster a better community with the larger threadiverse.
One issue a lot of anti-GenAI people keep bringing is that while they can block dedicated comms like !stable_diffusion_art@lemmy.dbzer0.com, they don't have an easy option to avoid GenAI content in random other /0 comms as there's no way to filter it out. This kind of content has been seen to cause a lot of strife, because people complain about its existence, while /0 admins and mods based on the above rule, tend to sanction those complaining. This then causes drama loops with /c/YPTB and /c/FuckAI etc.
There is a good point to be made here that while we don't mind GenAI content in /0, there isn't a reason to not help others avoid it. So we want to institute the following soft rule by now:
Simply tag your posts which consist of primarily GenAI content with the [GenAI] tag in their title. Not only will frontends like Tesseract will natively parse this as a tag and display it accordingly, but people who dislike such content, can simply filter it out of their feeds. Eventually lemmy will add tags which will make this tagging more seamless, but for now a manual tag in the title will suffice.
This rule only applies to posts in non-explicit GenAI comms. The assumption is that people can simply block those comms completely anyway.
As I said, this is a soft rule for now. Soft in the sense that you're not going to be sanctioned for forgetting it, but we hope people will remind you to do so. This is a good-faith attempt by us to co-exist and help others avoid what they don't want to stumble onto, much like [NSFW] tags. So I hope you'll add do a good faith attempt to help us in this. Furthermore, people who come to posts tagged as GenAI explicitly to scold and start slap-fights, will give the admins and easier justification to clean up, since they could have just filtered out that content in the first place.
Cheers
Most AI image generators that generate images add EXIF metadata indicating that the image is AI-generated. This helps people who want to identify AI-generated images readily.
In the case of ComfyUI, it even includes the entire workflow
like, another ComfyUI user can just grab the image, drop it onto their ComfyUI Web UI and they'll be right where the generating user was.
Unfortunately, EXIF metadata can contain location information
some cameras and such add it
and this metadata led to people posting images at places like Reddit being doxxed after they didn't realize that they were posting their GPS location and maybe real name, stuff that some cameras attach. As a result, a number of image-hosting places simply strip all metadata, to prevent users from from accidentally leaking this information.
Pict-rs, the software package that Lemmy hosts run to permit image uploads, does this. Unfortunately, it means that those "this is an AI-generated image" tags get stripped off.
So, for example, on my system, with ComfyUI, using ImageMagick:
"Properties:prompt" has a JSON encoding of the workflow.
Sample images generated by various AI image generators are readily-available on civitai.com.
For this generator that generated this image on civitai, it looks like the parameter is "Properties:parameters".
I believe that there are a small number of such tags today.
It would be technically possible to just not have pict-rs strip that particular tag (or tags, if there are multiple that a given generator adds?) off, have a list of "AI-generated tags", then have Lemmy add some visual indicator that an image is AI-generated. I'd suggest that this is probably a better longer-term route to indicate that an image is AI-generated than manually-tagging post titles, for a couple of reasons:
Spiders that index images on the Web will know that the image is AI-generated and can flag that for users and let them use that as a filtering criteria (e.g. Kagi Images permits for this). They aren't going to understand tags in post titles, but the metadata tags are somewhat universal.
Doesn't require manual effort if an image can have some indicator or flair or whatever put on it automatically. And I guarantee that some users are going to get this wrong just by accident, because different instances have different rules. Easier to change how a computer works than to change human behavior across-the-board.
Works on all instances.
The information remains attached to the image even if downloaded.
Works for images that aren't just the subject of single-image posts and don't have an associated title.
Speaking purely for myself, I kind of like the open-source, collaborative aspect of sharing the workflows or prompts, since it helps other users see how an image was created and learn from it; it's something that I'm glad to see the generators include, and I'm kind of sad that we strip it off on the Threadiverse.
exif is stripped when uploading to lemmy, I believe
I really appreciate this comment because I didn't know about EXIF metadata for AI-generated images and I think you raise interesting points about long term management of our informational ecosystems