stable-diffusion / AI - PART TWO

[Johnny Cage]

Originally Posted by Johnny Cage

Something odd, I can run five or six prompts with roop enabled in succession without problem. But then it invariably goes wrong. And any subsequent prompt just ignores the roop request. It only works again after I start up SD fully again.

Actually investigated more. It doesn't seem down to the amount of prompts I run. But what prompts. For some prompts SD just refuses to execute roop.

[BoInk2]

Originally Posted by Johnny Cage

The Checkpoint is just a big tits man. Can't really blame him.

Ok, got it.

Just wanted to do a Elsa Jean, and I liked her with her natural small titties.

[Johnny Cage]

Okay figured it out on my own.

Basically any prompt that it suspects is NSFW in nature will block roop.

We are aware of the potential ethical issues and have implemented measures to prevent the software from being used for inappropriate content, such as nudity.

The weird thing is that not all of the prompts that failed were NSFW. So I'm guessing it is detecting a lot of false positives as well.

Like it has refused fully clothed prompts but did allow a bikini shot?

[BoInk2]

Ok, we all like tits, so let's spice it up a bit.

[cylnz]

Originally Posted by lemonentry

Something like this? then yes it's possible

Like that, only she has to have brunette hair instead of blonde, naturally. The number 1 biggest issue I have with AI gen pics is tits. They almost never look like anything but bad silicone. And way too large. I suppose program training time will fix that in a few years.

How much control do you have on specifics? Does it work like a graph? Do you scale certain things like hair density by using a number scale? (hair density = scale .001 thru 1.0 ie. .001 = vin diesel and 1.0 = cher wig hair).

Right now I'm having a python3 PATH problem trying to get it to actually start.

thanks for the examples, how long did that take?

Originally Posted by Johnny Cage

Okay figured it out on my own.

Basically any prompt that it suspects is NSFW in nature will block roop.



The weird thing is that not all of the prompts that failed were NSFW. So I'm guessing it is detecting a lot of false positives as well.

Like it has refused fully clothed prompts but did allow a bikini shot?

So make an image in txt2img of anything then load that image in img2img NOT inpaint and then use a picture of your favourite starlet for face swap.

Originally Posted by cylnz

Like that, only she has to have brunette hair instead of blonde, naturally. The number 1 biggest issue I have with AI gen pics is tits. They almost never look like anything but bad silicone. And way too large. I suppose program training time will fix that in a few years.

How much control do you have on specifics? Does it work like a graph? Do you scale certain things like hair density by using a number scale? (hair density = scale .001 thru 1.0 ie. .001 = vin diesel and 1.0 = cher wig hair).

Right now I'm having a python3 PATH problem trying to get it to actually start.

thanks for the examples, how long did that take?

In a nutshell YES

http://andyhtu.com/post/prompt-weights-punctuations-how-to-use-it-in-automatic1111-stable-diffusion

Examples took about 10minutes

[Johnny Cage]

Originally Posted by lemonentry

So make an image in txt2img of anything then load that image in img2img NOT inpaint and then use a picture of your favourite starlet for face swap.

So it doesn't do the check for img2img?

Originally Posted by Johnny Cage

So it doesn't do the check for img2img?

Sometimes NO, you see the roop plug-in is a bit flaky as development was stopped a while back, so I heard/read.

I had to use inpaint and masking

From this.


To This.


Using This.

[Johnny Cage]

Can you recognize the porn star? And yes, my checkpoint also overdid the size of those cans.