ahdok
banner
socksandpuppets.com
ahdok
@socksandpuppets.com
I make comics sometimes https://linktr.ee/ahdok
I think incorporating them directly into the habit could be quite a cool look, because it'd give a strange silhouette... and historically some nun habits have been *crazy* shapes.

I've drawn a number of goblin designs over the years with the ears incorporated into the outfits, it's always fun.
November 15, 2025 at 10:23 AM
@pillarofgarbage.bsky.social has a video looking into this a bit, youtu.be/piZVmf29ju8 look at chapter-2 Asteria.
AI Is Not the Solution
YouTube video by Pillar of Garbage
youtu.be
November 15, 2025 at 9:57 AM
I think there's also some perhaps... shady? stuff going on here, a lot of parties seem invested in making an "ethical" poster child for the industry, even if it's just on paper and doesn't actually get used for any serious work, perhaps just to have something to point to in the face of criticism.
November 15, 2025 at 9:55 AM
The issue at heart with trying to make one of these is that licensing is expensive (assuming they did it properly!) and you're never going to have anywhere close to an order of magnitude of training data that the big players have if you're paying for it, so your model will always be lower quality.
November 15, 2025 at 9:54 AM
As for their models, they can produce very short clips that all have that airbrushed uncanny look and last only a few seconds each... not really "good enough" for any serious uses... (I probably should have specified that there are no "ethical" models that are high quality, not no "ethical" models.)
November 15, 2025 at 9:53 AM
After doing some digging in the last day, I have found a real company that claims to have a bespoke model trained "only on "licensed" video content", Asteria... However they're quite sparse on the details of what this content is or how they licensed it in the interviews I've seen.
November 15, 2025 at 9:51 AM
Or are the ear shapes built into the habit like arknights nuns?
November 15, 2025 at 2:12 AM
Would the habit have ear holes though?
November 15, 2025 at 2:10 AM
I'd argue there's a whole slate of other good reasons to not use AI tools (socio-economic, environmental, quality, ownership, legality, etc etc etc) - but it's a big sticking point for a lot of people.

Anyway, hope some of this is helpful :)
November 14, 2025 at 6:27 AM
Another way to think about this is- if a company DID train their "ethical Gen-AI" tool from scratch with entirely compensated/consensually offered content, you'd think they'd be pretty loud about that. A lot of people avoid these tools purely because of this one reason, it'd be a huge selling point.
November 14, 2025 at 6:24 AM
This is a couple years old, but it's pretty good at laying out the early history of how these models were built. It's a long watch:

youtu.be/-MUEXGaxFDA
The AI Revolution is Rotten to the Core
YouTube video by Jimmy McGee
youtu.be
November 14, 2025 at 6:21 AM
It's *theoretically* possible to train a model from scratch using only IP you own, but you'd have to own or license hundreds of millions of images, and then it'd take a decade of work and a few billion dollars. The models that are currently used started being worked on 20 years ago.
November 14, 2025 at 6:20 AM
When they were asked for more details, and specifically "did you train on existing works from other artists without permission?" they refused to comment.
November 14, 2025 at 6:18 AM
For example: Method Studio, (the people Marvel used to make the Secret Invasion opening,) were asked which tools they used, and their public statement was that they “skillfully leveraged the power of both existing and custom AI technologies to apply the otherworldly and alien look.”
November 14, 2025 at 6:17 AM
Often, after doing that, people will say "we trained this only on IP we own" - but what they're saying is that they started with something like Midjourney or Stable diffusion as a base model, then layered some "post-training" onto it with a specialized set of images, to make the final model.
November 14, 2025 at 6:13 AM
People often take existing models, then "train" a couple thousand images on top to make a model that outputs a particular style, and you can do that with a single artist's work (if they've produced a few hundred images or more), but there's still an underlying model you have to start with.
November 14, 2025 at 6:09 AM
I think it's possible to know... because afaik there aren't any "closed systems fed by a few consenting artists" that you can use to generate images - you need tens of millions of images to get a base model up and running at a level that's usable for anything.
November 14, 2025 at 6:08 AM
so he DID wrote a picture in his life
November 13, 2025 at 9:44 PM
I saw one where the photo on the pinboard was the one of them standing together, and the red string was like, 2 inches long... but I can't find it.
November 12, 2025 at 11:03 PM
Scully... do you believe in inference?
November 12, 2025 at 5:04 AM
When he says "racism is returning to politics" it's the one pledge he might actually stick to.
November 12, 2025 at 3:40 AM