AI Art Seething General

What are the chances it's gonna be another Resetera where within a few years the admin sells the website anyways to now owners who will then sell the data they accrued?
I don't think it'll get enough traction, most artists depend on normie SNs for traction, we're long past the days of deviantart and other such sites.
 
I wonder how long that site will last.
Cara's servers are melting because they're running glaze on everything that gets uploaded (which can be easily defeated via multiple methods, by the way.) So probably not very long.

What a wild pointless waste of time this is. You're gonna delete all your stuff off of FB and IG, which has already been fed into the machine, and then upload it to Cara where it will never be seen.

I don't expect artists to be that tech literate but like c'mon.
 
Recently, Meta (aka facebook, instagram and whatsapp) has announced they will use the data of their users to feed their new assistant AI.

Remember how many retards were claiming they would move to canada if Trump won the elections? Now they're doing the same thing, moving to pages like this one called cara. Which is the one i've seen being paraded around the most.
To me, it's funny seeing all these people trying to move away from the normie sites, you just know they'll get tired of seeing low engagement in their posts and will crawl back to zucc's feet for more likes.

I wonder how long that site will last.
>all of that shit is just ai slop or art of characters someone else made
-and nothing of value was lost. Doesn't matter how pretty and photorealistic your work is if you can't come up with your own ideas and stand by them.

Hence all the failing IP's these days by design.
 
Someone named Petravoice has gotten attention for this post.
1717307705919.png


Predictably this has caused noticeable controversy.
1717307668383.png


And then the following doomerism
1717307599417.png

 
How could these stupid idiots fall for this? Its so obvious its AI, the blurriness, the lack of detail, the odd mismatched colors, its not even pretending to be a real drawing, see this for example:
1717318005173.png
If I didn't tell you its AI would you be able to tell? be honest.
What's an SN?
Social networks, social media, all that crap.
 
If I didn't tell you its AI would you be able to tell? be honest.
Since the question is loaded, it looks like AI to me. Dunno. Based on parts of the reflection not making sense in relation to the scene above water, and some elements of the scene looking like near duplicates of each other (nearby branches of the trees stacked above each other etc). Also, the sun/moon being in front of the background treeline doesn't make sense. Could I have picked it out if I weren't already expecting it?

All of these are flaws (or just artistic license) that could potentially exist in a human-made pieces, too. It's nowhere near the obviousness of the earlier AI content with munted hands and such. Today, for a well-made AI image you need to know what to look for, and even then it's feasible there's genuinely indistinguishable images, especially in depictions and styles that are more interpretive and not realistic.

Since a model has the ability to regurgitate an almost exact image it was originally trained with under certain conditions, the potential for creating undetectable AI images is definitely there.
 
It is reasonable to believe that AI art is already absolutely everywhere. If you spend some effort on it, have some sense for aesthetics and are just somewhat competent at composition and photoshopping a little it's easy to fix the mistakes the AI might make in specific pictures. On good hardware, generating a picture takes a few seconds and there's absolutely nothing stopping you setting up a good workflow in e.g. comfyui, let the computer generate, let's say 500 pictures unattended and then just picking the cream of the crop in some thumbnail picture viewer. This is not even getting into sophisticated workflows, chaining different models and techniques together, with which you can do fairly advanced stuff, like have different models work to their strengths on one picture, do automatic impainting by having some other model recognize things in the pictuers automatically you want changed, have vision models automatically discard obviously screwed up pictures etc.. My dabbling in making loras a while ago before I focused more on LLMs made me notice that e.g. SDXL can be very competent at making pixel art. With some (automatic) post processing to reduce color palettes and actually turn it into even pixel art and touching up the mistakes, which is a very mechanic process with pixel art not needing much "expertise", it was easy to make pixel art that was impossible to peg as AI art. I am no artist, in any way. I could have spend time on it, uploaded that stuff to places and nobody would've been able to tell. I didn't because I had zero motivation to do so. I am sure that does not apply to everyone.

I'm pretty sure at least some of these artists shitting all over AI art on social media might secretly use it themsleves to make things. If you work to it's strengths, it can speed things up for you a lot. It's a very powerful tool if used well. Since many of them seem to refuse to learn the first thing of how these tools work, they are probably just familiar with straight txt2img generation on websites like OpenAIs Dall-E, which is really the most basic way to do stuff.
 
Last edited:
I'm pretty sure at least some of these artists shitting all over AI art on social media might secretly use it themsleves to make things. If you work to it's strengths, it can speed things up for you a lot.
I'd positively bet on it. And many of the rest who don't are just too dumb to figure it out.
 
Cara's servers are melting because they're running glaze on everything that gets uploaded (which can be easily defeated via multiple methods, by the way.) So probably not very long.

What a wild pointless waste of time this is. You're gonna delete all your stuff off of FB and IG, which has already been fed into the machine, and then upload it to Cara where it will never be seen.

I don't expect artists to be that tech literate but like c'mon.
IMG_20240602_130451_875.jpg
According to the site owner themself, just the server costs alone, are now at $2.200/month, while the site only makes money from Coffee donations. I don't think they'll sell out their users, but they'll have to find a more sustainable model, if they want the site to survive, as more artoids pour in.

The site itself is a nice blend of Artstation and Xitter, and I have to admit that the antiAI filter used, is not really noticeable (albeit mostly useless, especially against targeted scraping). I wonder if it butchers transparent PNGs and I think it is stupid, if it's impossible to opt out of it and have your art preserved 1:1 - reminding me of what I hated about old social media sites: using strong compression, bad resize algorithms, ...
 
I'd positively bet on it. And many of the rest who don't are just too dumb to figure it out.
I can give you one better. ArtPostAI is an artist, who had to get a day job working with AI and created this channel to cope with the existential crisis this got them into.

^ This is the most recent video, in the 3+ month long spiral.

Here is the second oldest video, that is more of a basic channel introduction:

Both videos above, include the AI job story.

If you want an example of a more concrete video from this channel, I recommend this one, as it is not only that, but also takes a bit of an unexpected turn. I expected it to be about the impossibility of producing pixelart, with diffusion models, but it is about everything else, except that:
 
How could these stupid idiots fall for this?
As someone who posted my art on Twitter, back when I started drawing stuff, I can tell you that nobody ever looks at your older art, or clicks your profile. They just follow and like the stuff you post afterwards, if the hecking algorithm shows it to them.

So I bet it's because the newest art looks way less AI generated and 99.999% of Twittards, never bothered to look at the older (more obviously AI) images, posted before that account blew up.
 
If you want an example of a more concrete video from this channel, I recommend this one, as it is not only that, but also takes a bit of an unexpected turn. I expected it to be about the impossibility of producing pixelart, with diffusion models, but it is about everything else, except that:
Just to clarify, AI is actually really good at large pixel-art pieces. The sort of stuff that Paul Robertson or Henk Nieborg would do.
Game assets are a mostly unexplored field and I imagine less success but I've been surprised before. It would likely require human massaging though.
If I didn't tell you its AI would you be able to tell? be honest.
If I examined it, I would probably be able to tell - however, in situations that it'll only be glanced at: It's gold.
Stock images are likely to become a dying breed.
 
My dabbling in making loras a while ago before I focused more on LLMs
What are you doing with LLMs?
SDXL can be very competent at making pixel art.
I'm not an expert on SD but the way the diffusion works it makes sense its good at pixel art.

Too bad that stability is going down before getting something on the level of sora going, that would be awesome to do character animations for 2D games, specially really fluid SNK-tier ones.
I'm pretty sure at least some of these artists shitting all over AI art on social media might secretly use it themsleves to make things.
Oh totally, would be interesting to hear from an old timer what was it like when photoshop and the like entered the scene, were the artists using it getting shitted on for that? I only know one guy who still draws old school by hand with pencils and brushes, everybody else has been using a wacom tablet or similar for over a decade, some I wouldn't be surprised if they never drew anything with physical tools since ipad kids are near 20 now.
just the server costs alone, are now at $2.200/month, while the site only makes money from Coffee donations. I don't think they'll sell out their users, but they'll have to find a more sustainable model, if they want the site to survive, as more artoids pour in.
I don't think there's much to sell out really.
So I bet it's because the newest art looks way less AI generated and 99.999% of Twittards, never bothered to look at the older (more obviously AI) images, posted before that account blew up.
The example given was supposedly her latest work and it still sucks ass, she truly doesn't understand how weights work, probably using a free online generator with limited options and low samples.

Frankly I think she's just popular because (if the right pic is her IRL) she's pretty. There are tons of mediocre artsy chicks who only get followers because they are hot, but their art is shit.
 
Just to clarify, AI is actually really good at large pixel-art pieces. The sort of stuff that Paul Robertson or Henk Nieborg would do.
Game assets are a mostly unexplored field and I imagine less success but I've been surprised before. It would likely require human massaging though.
In the gaming industry, so far it's mostly used for rapid prototyping. The moment the artists get asked to make NPC no.75 within deadlines, they hop in on the AI train after being introduced to wildcards. In the end, their creativity only goes so far, even with photobashing or 3d painting. Speaking of which, even the last 2 are contested subjects by artists sitting on social media all day.

If you're not slowly churning out pieces (Unless it's October), then you're not doing it right.
 
I wonder if it butchers transparent PNGs and I think it is stupid, if it's impossible to opt out of it and have your art preserved 1:1 - reminding me of what I hated about old social media sites: using strong compression, bad resize algorithms, ...
Glaze 2 does a better job of leaving transparent backgrounds alone but you'll still get weird little blobs.

Once you know what to look for you can tell if art has been run through Glaze and it actually gets annoying. It's the weird yellow-green disco lighting that's especially noticeable among single-color areas of the image.
 
Back