I am eagerly waiting for the day when let's say Koh-I-Noor will sell me an eraser, but force me to sign a contract so their (hilariously underpaid) employee can knock on my door any time, I am obliged to let them in, they'll shuffle through my family photos, comment on them being garbage and point out that I suck at art by the way, then grab a handful of my sketches, make photocopies of them and start selling them just down the street. Then drop an ad into my mailbox before leaving, just for good measure.
I got a Cara account back in October of 2023. I didn't really start using it until a couple weeks, though. I like it. It *is* slow and has some bugs that irk me a bit, but it's good as a whole. I like that it has stuff for you to enter information like the medium and genre of your art. Overall, it feels more artist-oriented than Instagram, which I really appreciate. Plus, you can also write text-only posts if you want (though I haven't yet).
on the glaze topic: it's an attack on the variational autoencoder (vae) of one specific ai model (stable diffusion). it's not a silver bullet against ai, and it's kind of trivial to counter. in simple terms, the way sd works is through an iterative, looping process of generating detail. to make this loop faster, the ai doesn't work on the full-resolution image, but rather on an ai-compressed image (commonly called the "latents"), which is about 50 times smaller than the full resolution image. the vae and its accompanying decoder is the specific part of the ai that compresses and decompresses the image, trying to stuff all the meaningful bits into that small of a space. glaze works by throwing off this encoder, and making it encode the image into nonsense. the ai is trained on encoded images, so if someone tries to train an ai on a glazed image, it is trained on nonsense. it achieves this through adding a simple smoke-looking pattern that's specifically designed to disrupt the ai's inner workings. the problems are: - this is trivial to detect (you can just run the image through an encode-decode cycle and see if it changed significantly) - it is also trivial to counter (some simple blur does it), the stable diffusion community did it within 30 _minutes_ after glaze was first released - the attack is specific to the ai model. a slightly modified autoencoder, a newer ai model that wasn't released yet when the image was published, or a closed-source model like midjourney or dall-e, are all unaffected by glaze - it's computationally intensive and therefore expensive to provide (which is why you only get to do it a limited number of times on this site) ultimately, it's likely just a burden on cara's hosting infrastructure without any real-world benefit except for marketing. i get that people want to do something but this is mostly snake oil.
I'd love to try the Mint Rollei, but when I got an RF70 it had to go back to Mint 3 times because of issues... The one I have still has some quality issues but they're livable with. Not ideal when the camera was so expensive to begin with for what it is, never mind adding hundreds of extra in shipping for unsuccessful repairs/replacements...
Hey Lou, Per your linked privacy article, it says US users cannot opt out of the AI post training function. The checkbox that European users see will not appear for US users. 😢
Yeah! Unfortunately we don't have the same setup as Europe-- this article also some info on how to object to it, I'm going to add it to the description too
your videos havent been showing up for a while, and i wondered what happened to you. turns out you have tons of videos and i just havent been seeing them in my sub feed! havent seen them in my rec feed either which is weird because i always interact with your videos so i should see them. is it just me these days or is the rec feed 90% videos that have already been viewed??
Lou, about the price for the Mint camera, compare with the prices of good digital cameras and it isn't that crazy. But i will still continue to use my 30 Minox 35 😊
Yeah that is a fair point as well!! But I already have my Pentax and stuff, so it's pretty tough for me to justify the 650$ (minimum) And I standby the idea that I don't think a beginner is shelling out that price either lol
