AI Art Seething General

  • 🐕 I am attempting to get the site runnning as fast as possible. If you are experiencing slow page load times, please report it.
Call me retarded if I am being. But man I fucking hate this model. I prefer the generic fat futa furry porn slop spam over this.
Now you see the reason why the creative kind of people gravitate on certain models. They want full control and customization over their image generations which allow for some crazy shit.

SDXL in particular is so popular there's entire base models derived from that and these models can be further customized with tons of derivatives sprouting out in that site alone.

Also I have yet to see Loras of the GPT1 model, wonder why 🤷‍♂️

Also bonus image:
Screenshot_20250711_075214_X.webp
 
Also I have yet to see Loras of the GPT1 model, wonder why
I'll just clarify, just in case, I am talking about a lora made for either SDXL or IL that was trained on exclusively images created using gpt1, not a lora made to be used on gpt1 if that's even possible.

There's a lot of them that are basically just memes. There's one guy that made a load of them. Bit hard to see but trust me when I say most of these are trained on gpt1.
rwegbbegrwrwdgvgrw - Copy.webp

Some of them are really good for doing the thing they're meant to do, but every single one of them will force inject that oil painting 'style' and fuck up the anatomy.

Glad I'm not just being retarded though. Sometimes I can get just compulsively contrarian, or just blindly hate things and paint them all with a broad brush just because I personally don't like them or find them reddity.

Also at least on civitai gpt1 is about 20x as expensive on the credits as sdxl or il and doesn't even fucking let me use loras so I can't even make anything with the character I started using AI for in the first place. I hate using (or lack thereof) the thing, I hate the 'humour' of the people that use it and I hate the people that make otherwise great loras that are enshittified due to being trained on shit.
 
I don't think OpenAI are smart enough to have engineered all of this, but they might have been.

- Recognizable piss filter is like a trademark, you can easily tell it's ChatGPT, and if the image is good despite the piss it acts as an advertisement to use their services
- Governments are happy because the images are always recognizable as AI, they might as well be watermarked
- People who care about quality won't train on piss images made by ChatGPT, so it prevents people from using their service to enrich competition, but conversely they can train on whatever they want

If it was truly an error in the way it was trained, they would've fixed it by now.
 
Call me retarded if I am being. But man I fucking hate this model. I prefer the generic fat futa furry porn slop spam over this.
While I had fun with gpt1 (had to prompt a lot to get used to it, and get around piss filter), I'm still sad how they butchered bing dall-e 3, and won't make a model like it. Yes, bing wasn't good at photorealism sometimes like gpt1 was or historical slop, but I preferred bing dall-e handled painterly styles. I do think a good portion is how they like each other's shit which gets pushed to the top, and gpt1, just really doesn't handle some painterly styles well.
_50ff478e-07a7-4d3f-8859-b7f3d18b0c5d.webp
 
i miss the old bing image creator too feels strange to say that i feel nostalgic for a fucking microsoft product but it was the first ai i properly fucked around with seeing what it can and cant do whereas now 99% of everything just gets dogged to the point that ive used it maybe a couple times this year just because most models are shit at making an image of my dog's breed and i wanted something dumb and quick to send to my mother it was the first time i saw the technology become something that you could get behind and would do what you said unlike the dallemini poc even used the images it made for a personal lora and yet nowadays its not even worth opening used to have it on my startup tab list and all
 
  • Feels
Reactions: Lovely Kin
faux oil painting, water color etc. is often wrong in execution. The biggest issue is that real ones don't have the level of detail image models often try to cram into them. Artists paint with intent and usually try to draw you towards something in the picture, and that's also where you can find the detail. Image models don't have intent and are all over the place with these styles. I think SD 1.5 based models were best with them, because their outputs were kinda less detailed in general. Also pretty much all newer gen models I know default to photorealism when they don't know how to pull off something in the prompt I've noticed.

The "piss filter" is probably a result of the pictures of oil paintings in the dataset having yellowed with age. I happen to own quite a few old oil paintings and they all look like this and also return to looking like this with time. Oil paintings sometimes have a varnish to protect them that has yellowed with age (a big one is that people used to smoke indoors a lot) and the colors used by the artist also change with time chemically, e.g. greens changing to brown, especially noticeable when the artist used layering for specific effect. You can have them restored but depending on the painting it's not cheap, it's only temporary and it can damage the painting especially when the restoration is too aggressive. The AI then applying the yellowing effect to pictures where it wouldn't be applicable (e.g. cartoon style) is just AI being AI. An example everyone knows is the Mona Lisa. It not only originally didn't look like this color-wise, she also used to have eyebrows that got lost because of over-eager restoration.

What I don't really get is why all people doing stuff with AI seem to be so fundamentally opposed to manually change the end result. A lot of the bigger AI mistakes in pictures easily go away with some minor photoshopping. It's like they themselves don't give these images any value worth investing work in.
 
Last edited:
I've had some really good results with dedicated watercolour loras honestly. And honestly most of the time if I make some image of a character the background is blank anyway. Could just be the lora I'm using and IL though? But I don't think the piss filter is a result of aged varnish. Ok yea almost all old paintings will have that discolouring but realistically how many old oil paintings are there compared to every other image in the training data? It's not just that it makes oil paintings look old, it makes every single image look like that, every gpt1 image has the same aged varnish oil painting look to it even when it is just completely not applicable. And honestly from watching those asmr satisfying porn whatever you want to call it restoration videos, I kinda prefer old paintings to have that old varnish look anyway.

And most people don't even bother with automatic changes let alone manual ones. I mean just from civitai forcing you to like 50 posts a day and seeing shit online, the vast majority of people are not upscaling images let alone worrying about denoising/highresfix/enhancing. Most people don't bother with photoshopping away minor details because they simply don't care. As much as I'll say yea ai based, a lot of the people using it are just lazy cunts using it to goon. The simple reality is that a technology that lets the computer do everything for you will always inherently attract people who want the computer to do everything for them.
 
The "piss filter" is probably a result of the pictures of oil paintings in the dataset having yellowed with age.
No, there's no way. it appears on all types of images, and these models are trained on billions of images. You would need hundreds of millions of images to be yellowed in order for the model to learn that everything that comes out of it needs to be yellow.

I have no idea how true this might be, but I read somewhere that they excluded one of the color channels from generation to save on cycles/gen time, which if that's even possible would make sense, considering the sheer amount of content they serve daily.
 
No, there's no way. it appears on all types of images, and these models are trained on billions of images. You would need hundreds of millions of images to be yellowed in order for the model to learn that everything that comes out of it needs to be yellow.

I have no idea how true this might be, but I read somewhere that they excluded one of the color channels from generation to save on cycles/gen time, which if that's even possible would make sense, considering the sheer amount of content they serve daily.
I think it's a temporary stop-gap. MidJourney doesn't have it.
 
The leftist anti-AI viewpoint is that old factories represent the decline of manufacturing jobs, and to them AI represents the decline of every other job, especially art.
Yea the anti ai luddites are now pro factory? Automation BAD!!!!!!!! that's why I am restoring this building that only exists to house automation? If a factory is a manufacturing job then working in mcdonalds makes you a qualified head chef really.
 
Yea the anti ai luddites are now pro factory? Automation BAD!!!!!!!! that's why I am restoring this building that only exists to house automation? If a factory is a manufacturing job then working in mcdonalds makes you a qualified head chef really.
I'm pretty sure they over-romanticize the realities of factory work, since they've never done it.
 
I'm pretty sure they over-romanticize the realities of factory work, since they've never done it.
Surely people online would never romaticise the olden days where everyone lived a fucking miserable life. No one would ever want to go back in time to a time period where your kids had such a high chance of dying before 5 years old that you didn't even bother naming them. Or when you'd get your hand crushed in some machine and they'd just come over and cut your entire arm off and you'd be left jobless and die living out on the streets because of it. Surely no one would ever be that retarded right? And if someone was that retarded surely they wouldn't be a politician and try to argue that the agricultural revolution was a bad thing and then cause a large portion of india to drop into famine because all the crops failed. Surely that would never happen right? Anyway the industrial revolution fucking sucks, even now factory work isn't much better. I'm off to go take part in my local roman and wwii reenactment group though brb.
 
The leftist anti-AI viewpoint is that old factories represent the decline of manufacturing jobs, and to them AI represents the decline of every other job, especially art.
I thought "Restoring Old Factory" meant he was renovating it for real estate or something?
 
It's virtue signaling.
But if AI art were to go away, at least the dA fanatics would be right back to nitpicking NI-made art for "art theft" again, just like they used to before AI art hit the scene. I still recall seeing those rants on dA.
Says "I want to be good at these things."
Oh yeah, and if AI-made art were to go away, anti-AI-art REE-ers would go right back to dumping on NI-made art they do not like as being "derivative" and "soulless" as well.
 
Oh yeah, and if AI-made art were to go away, anti-AI-art REE-ers would go right back to dumping on NI-made art they do not like as being "derivative" and "soulless" as well.
or idk, seeing anyone using digital art as cheaters who don't know the true pain of toxic paint fumes going past the nostrils. Can we just stop? All art is valid long as you know what you're doing. The issue are the folks who don't bother with fundies or misusing the tools set for them.
 
or idk, seeing anyone using digital art as cheaters who don't know the true pain of toxic paint fumes going past the nostrils.
That could've been a common sentiment back when digital art first started among artists used to traditional media. I think photography was seen as "not real art" when it was invented. All this arguing and changing values over what is "real art" or not is why I like the criteria I use to determine "real art", though it's best not to overthink it.

The way I see it, there's 3 ways art can be "real art", and the more criteria met the more "real art" it is. Of course, these criteria can still be subjective to an extent.

skillfully done
looks beautiful
thing is "deep"
 
Back