We are working on a new feature to self-label media...

All posts pixelfed's posts Post Back to profile

We are working on a new feature to self-label media as AI generated

In addition, we're building this into feeds and discovery so you can opt-out of AI generated media posts with a single click!

#pixelfed

Like 13 May at 10:04 | Open on mastodon.social

23 comments

Leonie :vf: :kopimi:

@pixelfed great idea. Mastodon should have that. https://github.com/mastodon/mastodon/issues/30277

13 May at 10:06 | Open on kopimi.space

pixelfed

One more thing, we'll make it easy for admins to enforce this by adding new reporting and moderation features to catch repeat offenders and monitor potential offenders with ease

We will get this done properly, and will be requesting your feedback once we ship the initial version ✨

13 May at 10:07 | Open on mastodon.social

Hirad

@pixelfed what does a potential offender mean?
Let me see if I got this right.
You'll consider posting an AI generated image, an offense that admin will be able to block users because of it. And somehow making it possible for admin to block someone even before posting an AI generated image as a potential offender?

I think you should explain it with more details.

13 May at 13:51 | Open on m.hirad.it

Emelia 👸🏻

@pixelfed how will these labels federate?

13 May at 10:07 | Open on hachyderm.io

[DATA EXPUNGED]

Emelia 👸🏻

@pixelfed definitely curious in review it, though I wonder if you can just use tags on the media?

Like: Note -> Attachment -> Tag and use Hashtags on the media itself, in parallel with Tag on the Note itself

13 May at 10:57 | Open on hachyderm.io

Roundcat

@pixelfed wow, a built in art theft detector!

13 May at 11:37 | Open on mastodon.social

Bohdan Pavuk

@pixelfed will this be part of #activitypub protocol?

13 May at 13:03 | Open on mastodon.social

pinkdrunkenelephants

@pixelfed You need to ban it entirely. Bare minimum make AI opt-in, not opt-out.

13 May at 13:15 | Open on kolektiva.social

gregandcin :youtube:

@pixelfed Love the work @dansup
and contributors are putting into the tools ecosystem for ActivityPub!

13 May at 13:20 | Open on mindly.social

Hirad

@pixelfed and how is it gonna detect an image being AI generated exactly?

13 May at 13:42 | Open on m.hirad.it

VOLTΛIRE

@hirad @pixelfed they didn’t say anything about detecting, just self labeling

13 May at 16:04 | Open on chaosfem.tw

Hirad

@2d @pixelfed
Except in 2nd post they speak about catching repeat offenders. If it's just "self-labeling", then what is there to catch exactly? What is the offense?

13 May at 16:08 | Open on m.hirad.it

VOLTΛIRE

@hirad @pixelfed then that would be people who post AI content and label it as not ai

13 May at 16:09 | Open on chaosfem.tw

Hirad

@2d @pixelfed and how is that gonna be done? The AI content must be detected somehow to be able to judge if a post should have been labeled or not.

13 May at 16:12 | Open on m.hirad.it

VOLTΛIRE

@hirad @pixelfed usually people just look at it

13 May at 16:26 | Open on chaosfem.tw

Hirad

@2d @pixelfed I'm speechless.

13 May at 16:43 | Open on m.hirad.it

VOLTΛIRE

@hirad @pixelfed do you think an algorithm would be more effective than a human moderator?

13 May at 17:41 | Open on chaosfem.tw

Hirad

@2d @pixelfed
I don't even think admin is entitled to decide about that. As long as an image is not illegal, insulting or inappropriate (you know, the regular rules instances have), it shouldn't matter if it's a picture taken by a camera, painted, drawn, generated by AI or produced in any other way.
Generating AI image is not an "offense" to be caught and moderated beyond standard moderation rules.

13 May at 18:04 | Open on m.hirad.it

VOLTΛIRE

@hirad @pixelfed oh but it completely could be under this system - lying about your image not being ai generated could be very frustrating on, for example, an art or photography focused instance.
I’m not as hardline anti AI as many on mastodon, but there are reasons why people might not want to see ai content, and if people break rules to get around that, it can be a moderated action.

13 May at 18:09 | Open on chaosfem.tw

Hirad replied to VOLTΛIRE

@2d @pixelfed
That's a good point. I hadn't considered those instances.
Will see how exactly it's gonna be implemented, but I think I agree with you now that it can be beneficial.