Now that #ArtStation has finally done *something*, namely adding the "noai" meta tag to the header of HTML documents if you opt-out. let's talk about why it doesn't actually DO anything to combat Image Generators.
For example:
1. Common Crawl scrapes the internet, ignores HTML meta tags
2. LAION sorts that data into URL +Metadata pairs.
3. img2dataset downloads the images with the URLs, but doesn't even check the HTML for meta tags.
4. The model is trained with your art.
#Ai #CreateDontScrape
@Curator Hey Curator?
Do you have control over the robots.txt file of your instance?
I've recently learned that ALL of the data that Stability Diffusion uses, comes from Common Crawl, which scrapes the entire internet once a month.
Since so many people here use Alt Text, it's a hotbed of well-described image-text pairs (Juicy data for the AI evangelists).
Is it possible for you to block it?
So littering on land is not acceptable, but apparently killing turtles is fine? 😅
Death of a culture
The death of a culture~
Somewhere the vulture
Somewhere the vulture~
Is waiting for us, the apartheid son
The son with a twitter gun~
Evil comes in Musk shapes
I've been looking for something exactly like this! Maybe I should just pull up my sleeves and make it myself bespoke for my game, since I probably have some slightly different needs. 😅
Adding a small image with alt-text.
Fellow Mastodon-Verified folks!
When adding the Mastodon link to your website for verification, consider adding a meta tag with
name="robots" content="noimageai, noai"
to send a message to webcrawlers that you do not want your images to be harvested for AI datasets.
This is not an "official" directive yet, but Deviantart just started using it. Might help for a potential class-action lawsuit in the future. 🤷♂️