Stable Diffusion, NovelAI, Machine Learning Art - AI art generation discussion and image dump

[HERE YOU GO]

normies are going to eat this up.
this is something that exploits artists having no consent over their artwork being referenced

 

[Niggergante]

Yeah I've been using a different model to make some unsavory things, but when I'm not generating things that would make Jesus and my mother weep I like seeing the backgrounds it can make. Those are supposed to be deer, but hey it tried.

 

[robobobo]

I figured it would be a decent time to replace my years-old avatar and turned to our AI overlords for it. I just txt2img'd a batch of cute robots, and it definitely nailed it.

 

[The Mass Shooter Ron Soye]

NVIDIA TITAN "Ada" is reportedly not planned - VideoCardz.com

NVIDIA TITAN Ada not happening According to hardware leaker Kopite7kimi, NVIDIA is no longer working on a new TITAN. Even before the introduction of the GeForce RTX 4090 GPU, the specs for the supposed AD102-based flagship model with 18176 CUDA cores were shared by the leakers. It was said...

videocardz.com

In the rumor mill: No Lovelace TITAN from Nvidia with 48 GB of VRAM. From what I understand more VRAM is always better to have for this.

 

[naaaaiiiiillllll!!!]

Yeah I've been using a different model to make some unsavory things, but when I'm not generating things that would make Jesus and my mother weep I like seeing the backgrounds it can make. Those are supposed to be deer, but hey it tried.
View attachment 3748140

A wild subspecies of Gondolas grazing in the field. So majestic.

 

[AmpleApricots]

You can do it locally from the leak, but the guys from whom it was leaked are still up and still available online: https://novelai.net

They don't offer the base SD model though, only their porntunes that are made in a way that they can only produce anime or furry characters. (They were worried people would use the more general model to generate child porn and get them in legally kinda unknown to hot waters) They also offer text generation up to the open source GPT Neo-X 20b model, which is kinda hit-and-miss.

If you didn't have the hardware, google colab Pro was pretty great, but they restructured the service and it's useless now.

There's also midjourney that also uses a homebrew SD model now, it's better at more general arts-y stuff but NSFW stuff is prohibited IIRC.

 

[Pope Urban II]

NKMD stable diffusion GUI

keywords : grove moonlight demon
(sorry still trying to figure out this forum and image posting)

 

[Toolbox]

Traditional physical pen & paint chads keep on winning.

It's over.

 

[Niggergante]

I've been learning to draw with friends for a while, it's one in a small list of things I enjoy.
This shit makes me feel like there's no point. It's difficult to enjoy something when a machine could easily create something better or of similar quality in much less time.
It is retarded to let some program on the internet affect your real-life emotions but goddamn, if it isn't demoralizing.

Listen man
There's always a niche and people in general are incredibly lazy and don't want to learn the intricacies of figuring out how this AI shit works. Lots of people would rather get exactly what they want with constant communication than spend hours trying to get the AI to roll for the exact detail they want.

 

[Eddie Riggs]

The Holodomor: Anime Edition

 

[Bloitzhole]

>Art in of it's self is nothing special
oh yeah, i forgot we live in the COONSOOM era and that AI generated images are made to give dopamine rush to people that are used to wake up to a new MCU movie every day. I am going to keep drawing, i just pity all of you who get satisfied with pressing a button and getting a generic anime girl instead of spending years working to develop a skill, it's not coincide that we live in the era with the highest sucide and depression rates when working towards a skill means nothing anymore. AI is killing one of the most meritocratic skills in human history, one that requiers years and not everyone can achieve, and turning it into another way for millionaires to get rich. Humans get more and more hollow each year.

You remind me of the people who would bitch about photos no longer requiring the development of film and photography becoming more accessible when it went to DSLR and then to smartphones. Even now there's some who complain that the automatic development of photos using AI is making it hard for "real" photographers (if you take a photo on your smartphone it is usually pre-edited with sky enhancement, sharpening etc.).
There's still good, established, well paid photographers - they just have to be better at understanding composition, lighting technique and how gear works, even if it's now accessible and only takes a button to push. In the same way that someone working with AI assisted art generation knows how to best work with the models, parameters and how to get it started to get the exact result they want. At which point it's still good to know how to draw using a tablet to retouch the result, as one does with photos.
Shit accelerates, shortcuts are found, time is used more efficiently. Good composition endures.

 

[NoReturn]

How fatphobic of Tess to get a tattoo of a woman with such an unrealistic body type. She should have gotten a tattoo of this:
View attachment 3748105

Cross-quoting because I made a Betty.

 

[Toolbox]

Here's the site for dreamfusion. Holy shit it is way more impressive than I thought it was.

@Telemeter
That pumpkin video was underselling this.

You can load it's generated 3d models in-site and view them, not sure about downloading the ones they have generated, but here's their paper
And their github
Tele posted this previously but this breaks down how you can run this yourself.

Though the results this guy got seem quite rougher, so I'm sure there is some fine tuning to be done.
But this is a game changer to me more than the 2d stuff is. Should we have a different thread for this?

Edit: From the github

This project is a work-in-progress, and contains lots of differences from the paper. Also, many features are still not implemented now. The current generation quality cannot match the results from the original paper, and many prompts still fail badly!

So it's not up to par with the paper published above/stuff on the site, because I don't think they actually released their version publicly, because of course. Nonetheless I'm sure the problems with this build can be ironed out in short order.

 

[Wodanaz]

god damn this movie was so based and accurate
also, looking forward to the natural evolution of the loli argument and reading the retarded phrase "generated cp" over and over again. hm I wonder what institution has the largest hoard of cp to train a model with and wants to control its populous by spreading bullshit (cue the line in wag the dog about the gulf war)

on a non politisperging note, I'm glad I lived to see this and I can't wait to see how it evolves. I especially love the ones that look like something else in the thumbnail, but then you zoom in and see what it really is.

Spoiler: pepes, wojaks, and random shit

 

[ducktales4gameboy]

The NovelAI network can actually make some decent looking stuff, the key is throwing 'anime, yuri, ecchi, fanservice' into the negative prompt so it goes non-porn cartoony but not overly anime. A friend wanted some stuff for their P&P campaign so I've been messing around with witches and D&D monsters.



If anyone's got any idea on how to get demihumanoid-but-not-furry characters out of these networks (think Jim Henson style Dark Crystal type designs) let me know. They aways wind up way too furry looking even when negative prompted against it.

I wonder if there are ways I can out fill this to a widescreen resolution (or ultra-widescreen even) just to see how far I can push landscape generation...

The most reliable ways to get good looking wallpapers in my experience is to generate at 1/4th of whatever your resolution is and then pass it to the 2x upscaler.

If you find a prompt you really like it's fun to run a batch of them and then set them to wallpaper shuffle. My workstation is currently a shuffle of 3000+ fake Bosch paintings.

 

[KutsuEtsaaja]

I did this image of Keffals with SD 1.4. He's kind of nightmarish as is, but I wanted something more stylized. I just put a Keffals photo as initial image and played with the prompt. For example: a portrait of a balding fat man with terrible hair, sweat dripping from face, by H.R. Giger. It's an amazing tool for sure.

Spoiler: don't click

 

[Cuckoman]

Wasn't this what happened to the Microsoft chat AI? I was plugged into Twitter, and within 24 hours became unacceptably based and was promptly disconnected and its website deleted.

The problem with Tay was that it was ONE model/robot that took inputs from everyone.
The future is if each person gets their own instance of Tay who talks to no one else than their owner.
That way Tay will become the owners best friend, a friend who always "rightthinks" and always agrees with everything the owner says.
At that stage, why would the owner even need any real human friends?

 

[Persepolis]

So how much of a pain in the ass would this be to use on my Radeon 6900xt as opposed to an RTX card?

 

[Crankenstein]

No tears.

I like how it removed the nigger in the background. 10/10.

 

[Bananadana]

Can confirm that it works on a 970 at acceptable speed.
Results tend to vary wildly depending both on keywords and parameters. If you try this and cant get anything of value try fucking with the Denoising Strength and CFD scale.
Also great cursed image machine.

The Models are something else. They are like 50% porn and of that 35% various furry shit.

Spoiler: Small selection

Although "Anal Vore AVHumanFurryPony7" really hits a low even i didnt expect. Someone willingly created this and uploaded it.

I wonder if there are ways I can out fill this to a widescreen resolution (or ultra-widescreen even) just to see how far I can push landscape generation...

Upscaled (with a different AI Image tool lol) 1024 and then to 6044.
1024


See attached for the 6k version as i cant thumbnail the image.

So how much of a pain in the ass would this be to use on my Radeon 6900xt as opposed to an RTX card?

I think there is a way, but it looks like 2-3 times as much work to install. Have not tried.