These Women Do Not Exist ..... The NSFW A.I. Art Open Thread

[D24OHA]

Seems like they have a lot of trouble replicating hands & feet. Why is that?

I would assume it's because in reference images, hands/ feet are not always prominent or in focus.....

i.e..... the image is focused on the face/bust of a person.

Also people can have their hands in soooo many different positions. So the lightning/shading can be weird. Since AI isn't actually "seeing" the images it's replicating, the weird shading/lighting/positioning is what comes out in the renders.

This article states, ai doesn't really "know" what a hand is to be able to manipulate it and render it all the various scenarios. It just mixes a cluster of reference images and voila..... however I think that is the very simple answer to a very complicated issue/process

Why AI-generated hands are the stuff of nightmares, explained by a scientist

Whether it's hands with seven fingers or extra long palms, AI just can’t seem to get it right.

www.sciencefocus.com

 

[Mixd]

I would assume it's because in reference images, hands/ feet are not always prominent or in focus.....

i.e..... the image is focused on the face/bust of a person.

Also people can have their hands in soooo many different positions. So the lightning/shading can be weird. Since AI isn't actually "seeing" the images it's replicating, the weird shading/lighting/positioning is what comes out in the renders.

This article states, ai doesn't really "know" what a hand is to be able to manipulate it and render it all the various scenarios. It just mixes a cluster of reference images and voila..... however I think that is the very simple answer to a very complicated issue/process

Why AI-generated hands are the stuff of nightmares, explained by a scientist

Whether it's hands with seven fingers or extra long palms, AI just can’t seem to get it right.

www.sciencefocus.com

So far I'm seeing there's LoRA's that corrects them and feet. Or also SDXL checkpoints handles them better with lower steps.

Sometimes the more steps and the higher CFG, the AI makes things sometimes worse. Depends on the model

 

[The Plutonian]

Need to learn this soon

 

[Duece]

 

[Jake Reed]

SDXL Prompt: naked, breasts visible, nudity, 33 year old, mulatto woman

I still can't get over only having to put "mulatto" in the prompt now. It's rare for the image to turn out like this on SD1.5.

I didn't think I could run Fooocus with just 6GB of VRAM but it runs well. SDXL takes me the same amount of time to render an image as SD1.5 given all the extra steps I added to the workflow (custom highres fix).

 

[no2plex]

Need to the titties out on this one fam!

 

[Mixd]

SDXL Prompt: naked, breasts visible, nudity, 33 year old, mulatto woman

I still can't get over only having to put "mulatto" in the prompt now. It's rare for the image to turn out like this on SD1.5.

I didn't think I could run Fooocus with just 6GB of VRAM but it runs well. SDXL takes me the same amount of time to render an image as SD1.5 given all the extra steps I added to the workflow (custom highres fix).

It runs on 4gb of vram and 8gb of ram

But for different races, that's more in the checkpoint, what was baked in. I've been using Afrodite, NewReality and JibMix. Fooocus I read was built off of Juggernaut, and remember it uses 60 steps.

There's also in Fooocus a thing called wildcards. You can create a text file with different things on each line. Meaning I have one called "races" I got from Civitai. You just put "__races__" in the prompt like you did mulatto and it will mix it up.

So you can create your own wildcard and condense the different races you want to use to variate the faces or types of race. Like I did with the sweater thing above.

There's a few different wildcards for colors, places, all kinds of things.

This is the one I got

Races + Mixes - Mixed Asian Races | Other Wildcards | Civitai

Dancing where angels fear to tread, I've generated a list of racial groups including mixed races, for example: Chinese-Malay, Afro-Latino, etc. Thi...

civitai.com

They're just text docs, so they're like 1kb

 

[Jake Reed]

It runs on 4gb of vram and 8gb of ram

But for different races, that's more in the checkpoint, what was baked in. I've been using Afrodite, NewReality and JibMix. Fooocus I read was built off of Juggernaut, and remember it uses 60 steps.

There's also in Fooocus a thing called wildcards. You can create a text file with different things on each line. Meaning I have one called "races" I got from Civitai. You just put "__races__" in the prompt like you did mulatto and it will mix it up.

So you can create your own wildcard and condense the different races you want to use to variate the faces or types of race. Like I did with the sweater thing above.

There's a few different wildcards for colors, places, all kinds of things.

This is the one I got

Races + Mixes - Mixed Asian Races | Other Wildcards | Civitai

Dancing where angels fear to tread, I've generated a list of racial groups including mixed races, for example: Chinese-Malay, Afro-Latino, etc. Thi...

civitai.com

They're just text docs, so they're like 1kb

Thank you, I installed a wildcard extension in Automatic1111, but had no idea that it was built in to Fooocus. In terms of racial diversity, I admit to being a little color struck, but it didn't help that the model (lifelike) I was using wasn't creating beautiful darker skinned women. I put one of them up here and got a negative response a few months back. So I did have a focus on trying to get them to come out lighter, but pumping out Cari Champion or prime Kenya Moore level dark skin women consistently would be ideal.

I'm going to play around with it and see what it can do, but that "mulatto" is a cheat code basically.

 

[Mixd]

Thank you, I installed a wildcard extension in Automatic1111, but had no idea that it was built in to Fooocus. In terms of racial diversity, I admit to being a little color struck, but it didn't help that the model (lifelike) I was using wasn't creating beautiful darker skinned women. I put one of them up here and got a negative response a few months back. So I did have a focus on trying to get them to come out lighter, but pumping out Cari Champion or prime Kenya Moore level dark skin women consistently would be ideal.

I'm going to play around with it and see what it can do, but that "mulatto" is a cheat code basically.

Turn on "dark fantasy" and maybe take off some of the out of the box settings. Look at what some others use for their outputs. But also the same guy that posted that wildcard, look at his others, he has one for "skin tones"

 

[R-One]

 

[R-One]

 

[Mixd]

Thank you, I installed a wildcard extension in Automatic1111, but had no idea that it was built in to Fooocus. In terms of racial diversity, I admit to being a little color struck, but it didn't help that the model (lifelike) I was using wasn't creating beautiful darker skinned women. I put one of them up here and got a negative response a few months back. So I did have a focus on trying to get them to come out lighter, but pumping out Cari Champion or prime Kenya Moore level dark skin women consistently would be ideal.

I'm going to play around with it and see what it can do, but that "mulatto" is a cheat code basically.

Check this for more realistic look, going to mess with some prompts trying this out

Fooocus Advanced · lllyasviel Fooocus · Discussion #117

Fooocus has automated the best configuration by default so that users can focus on prompting and generating, and ignore these parameters. Nevertheless, we also provide a bunch of features for advan...

github.com

 

[Duece]

 

[R-One]

 

[Mixd]

 

[Mixd]

No nudes but showing skin detail in Fooocus

 

[RoomService]

AI poisoning tool Nightshade received 250,000 downloads in 5 days: ‘beyond anything we imagined’​

How Nightshade works and why it is popular​

Nightshade seeks to “poison” generative AI image models by altering artworks posted to the web, or “shading” them on a pixel level, so that they appear to a machine learning (ML) algorithm to contain entirely different content — a purse instead of a cow, let’s say. Trained on a few “shaded” images scraped from the web, an AI algorithm can begin to generate erroneous imagery from what a user prompts or asks.

On the Nightshade project page, Zhao and his colleagues — Shawn Shan, Wenxin Ding, Josephine Passananti, and Heather Zheng — stated they developed and released the tool to “increase the cost of training on unlicensed data, such that licensing images from their creators becomes a viable alternative.

https://venturebeat.com/ai/ai-poisoning-tool-nightshade-received-250000-downloads-in-5-days-beyond-anything-we-imagined/

 

[Mixd]

 

[Jake Reed]

Check this for more realistic look, going to mess with some prompts trying this out

Fooocus Advanced · lllyasviel Fooocus · Discussion #117

Fooocus has automated the best configuration by default so that users can focus on prompting and generating, and ignore these parameters. Nevertheless, we also provide a bunch of features for advan...

github.com

I'm curious about this image prompt tab. I want to see if I can use it to get a consistent face or have it influence the subject to give me a similar look.

machine learning (ML) algorithm to contain entirely different content — a purse instead of a cow, let’s say. Trained on a few “shaded” images scraped from the web, an AI algorithm can begin to generate erroneous imagery from what a user prompts or asks.

On the Nightshade project page, Zhao and his colleagues — Shawn Shan, Wenxin Ding, Josephine Passananti, and Heather Zheng — stated they developed and released the tool to “increase the cost of training on unlicensed data, such that licensing images from their creators becomes a viable alternative.

Artists will still complain despite knowing for a fact that no ai art generators are being trained on their artwork. The problem for them is that ai art takes eyes off their work. It's art inflation.

 

[Mixd]

I'm curious about this image prompt tab. I want to see if I can use it to get a consistent face or have it influence the subject to give me a similar look.

Don't use the 1.5 models in Fooocus. I kept getting the same face/girl. I learned that old models don't work in it, just SDXL.

When I used the 1.5 as a refiner, it kept making one face and turning chicks white, a different race.

And if you want to keep the same face, just try using the same seed

 

[Mixd]

 

[Mixd]

 

[Mixd]

 

[Mixd]

 

[R-One]

 

[R-One]

 

[Mixd]

For those who don't have a graphics card and know how to change some lines of code in python (can do it quickly in Notepad), this dude got Fooocus to run on an integrated GPU and it runs pretty quick. It's about 2 mins for a 1024x1024 image which is is pretty good for not having a card.

Thank you so much for the latest release and all the hard work your team has done! · Issue #2116 · lllyasviel/Fooocus

First let me thank you @lllyasviel and the whole dev team 4 this great product you have made. I have tested and compared nearly all Stable Diffusion products out there on the open source market for...

github.com

 

[R-One]

 

[Mixd]

​

 

[Mixd]

 

[Mixd]

​

 

[Mixd]

 

[R-One]

 

[Jake Reed]

Don't use the 1.5 models in Fooocus. I kept getting the same face/girl. I learned that old models don't work in it, just SDXL.

When I used the 1.5 as a refiner, it kept making one face and turning chicks white, a different race.

And if you want to keep the same face, just try using the same seed

Yeah, I had the same issue with it turning the women asian/white using 1.5 as a refiner. I'm assuming I can use any SDXL model as a refiner.

In the image prompt tab there is something in there you can select to have it give you consistent face.

 

[Mixd]

Yeah, I had the same issue with it turning the women asian/white using 1.5 as a refiner. I'm assuming I can use any SDXL model as a refiner.

In the image prompt tab there is something in there you can select to have it give you consistent face.

I learned you really don't need a refiner with SDXL though. I was just looking at an old post when I shared that, dated back in Oct.

Messaging this other poster on Civitai, he was messing with Serafino which is more an anime model and then using NewReality as refiner. It's what I used above for the underwater images. Output is very realistic