PieFed 1.7.12 detects AI generated images
OpenAI, Gemini and others have gotten together and agreed on a standard way to embed metadata in images so platforms can tell if an image was made with AI or not. It's called c2pa.
PieFed now uses c2pa to label image posts as AI, where possible. This will not be 100% reliable because often as images are passed around the embedded metadata gets wiped out but it will help reduce the moderation burden involved with labeling AI images.
If you don't want to see any AI generated content, go to https://piefed.social/user/settings/filters and set the "AI-generated posts" setting accordingly.
If you are posting content that you know is AI generated, in the post creation form click 'More options' and then find the "AI generated" checkbox. If you're creating an image post and the image file contains c2pa metadata (i.e. you downloaded it straight out of the LLM without further modifying it in GIMP, etc) you don't need to bother - PieFed will take care of it when you submit the form.
When an image post from Lemmy arrives PieFed will attempt to read the c2pa metadata but there's a good chance Lemmy will have stripped it out while resizing the image. Image posts coming from other PieFed instances will already have been checked and arrive pre-labelled.
Also another change in this release is emoji reactions are more prominent and can be made from the teaser view, without going into the post itself. I hope this leads to more reactions being made and we all get a more interesting and vibrant feed as a result.
https://codeberg.org/rimu/pyfedi/releases/tag/v1.7.12.1Open linkView original on piefed.social
8 replies
It makes sense that emojis are controlled similar to votes, since they're basically a type of vote.
Yes I'd love it if people only voted on posts they have taken the time to fully read but that ship has sailed so we might as well do the same for reactions.
I feel like the emoji reaction is my knee jerk feeling, then I open the post/read/vote accordingly. Like for this post my initial reaction was the little party guy because that’s such a cool feature! And the general content+context of course gets it the upvote. :)
FYI under the hood, all reactions are recorded as an upvote. No need for both.
Could something similar be done with SynthID?
Not yet, there is no publicity released code by the powers that be so we can do it locally, only some sketchy reverse engineered stuff that might not work. At present you need to use googles API to check a Synthid watermark, which is far from ideal.
@[email protected] @[email protected] c2pa is for a whole lot more than just AI detection, it's full provenance embedded — stream.place uses it iirc for adding all the licensing data to live streams
This is awesome. It may be surprising but im not going to block ai content. Surprising in that I constantly talk about blocking and filtering so a user can have a good feed. I don't mind ai and I block meme communities because ai or not im looking for something else in my feed. That being said if someone makes a nice image of the pnp rpg character they imaginged that might be cool and such.