General AI Picture Thread

And more

Almost of them

Last of these

Wedding

I like the concepts of those but I think a lot of them have "bad anatomy" and maybe a few more iterations would have been able to fix this. Or maybe img2img even?

May I ask how you created those and which checkpoints you used?

bravo cDupin and gregmartin! Love to learn more about doing this...

I am always happy to help. Have you tried setting it up yourself? What are you struggling with?

I really like the bride and groom series in #34. What a great way to start a marriage!

It all seems so complicated I would love to do it, also can you use your own pictures

It all seems so complicated also can you use your own pictures

It's really not that complicated, though. I know it looks a bit daunting but if you follow the step by step guide it's easy to set up. And once you got a model going it's not that hard, it "just" takes patience.

Is there anything in particular that you are struggling with? I'm happy to help.


I am not sure what you mean by that. I assume you mean taking a picture and "ai-ify" that? Yes, that's possible. Depending on what you want to do the problem is that facial features tend to get lost and the people look very generic. If you use a picture just do do a certain pose for example, it can be helpful.

Here is an example, I am using a picture that is quite famous I think. You can use that as a base image and then "ai-ify" it. The stronger the "denoising strength" the more it ignores the original image - or in other words: the lower the strength, the more it keeps to the original. Here are a some examples with the denoising stength increasing. As you can see, her face gets more and more generic.

You can use all your models and therefore create different looks, depending on your model.

And you don't "just" "ai-ify", you can change things, depending on what you prompt. The "crazier" your changes, the stronger you must denoise and the less the person looks like the original.

Depending on what is important to you and how much patience you have, you can try to get it right. For example: I didn't prompt anything about the background, if that was important to me, I could include something in the prompt. Same about facial expressions. And obviously you have unlimited tries to get everything right.

I hope that answers the question and it's illustrative what I mean.

thats really damn hot
.
Submissive Cuckold - lives for many years in a female-led marriage with a cuckold lifestyle.

What programmes to i have to download to start trying and i was gonna do a picture of me with a big blk man doing me from behind and send it to my husband

You need to download stable diffusion. There is a step by step guide on page 1 of this thread, though I recommend using a chatbot like perplexity.ai to guide you through it (you can ask it if you encounter any errors for example).


That is certainly possible, though as I said, you will probably look less like yourself because in order to insert somone you have to denoise guite a bit and your face (or whatever you show of yourself) will become more generic. You could also do "inpaint" and tell the program to insert someone behind you and leave you (or parts of you) the way you are.

If you want to you could send me a picture (that you are comfortable sharing) and I could give it a try.

cDupin
Yes, I was looking to use existing photos to generate some AI pics. I want to use some hot pics of my wife (she is a hotwife) yet not disclose her face because of privacy issues.

By the way, I love how hot and real these images can be - thanks for posting!

All right, makes sense. So where in the process are you? Have you got stable diffusion up and running and wonder how it works or are you just thinking about doing it?

As I have shown (or tried to at least) it is possible though you have the issue of people becoming generic. You can tell the AI to change everything but the face or only do the face a little bit and do the rest more. I had some decent results with that - for privacy issues nothing I want to show here but I might do some examples later on if I have time and people are interested.

As I have said before, it is also possible to create a model on your own photos which would give you better results, I haven't looked into how this is done though, I just know it's possible.

To give you a better understanding of what I mean, here as an img2img explanation I gave to someone - it was specifially for improving images you generate with AI but the same principles apply if you use a "real" image. I hope it helps you all understand how it works.

One use case could be that you are very happy with your image, everything is perfect, you just wish one detail would be different. Maybe you want the girl in the shot to have cum all over her, than you could load her up in img2img and change the prompt, and add something like "cum on breasts" or whatever. Than you add the picture, change the prompt et voilá, you don't have to churn out pic after pic but just "alter" the one you already have

Let's say you wanted the cumshot to be specifically on her breasts but the img2img made it on her face and crotch, too. You could use the inpaint function and "paint" over the picture. The new prompt will then only do changes on the parts that you have designated i.e. painted over

and like i've written. Low denoise keeps the picture pretty much intact/unchanged, higher values will do more extreme changes. Depending on what you do you should play around with it. If the changes are to big, it might be easier to do new ones altogether. It can also be helpful to do some photoshop action, add something to original, then plop in the photoshopped version and then do img2img as this will help the AI do what you have in mind (I hope you understand what I mean).

cDupin

How do you manage to keep the same faces consistent throughout the comic? I also struggle to create scenes with more than one person.

I'd say everybody struggles wit those issues. It's one of the shortcomings of AI image generation.

For consistency my "secret" is that I use very generic characters so the AI doesn't have a hard time reproducing them. I try to make them recognizable by things like hair colour and cut or give them one aspect that stands out. Other than that I can only recommend to keep the prompts simple.

In regards to several people: for me it helps to write what you want the characters to do first (as the beginning of a prompt is most important for the ai) and do character descriptions later. So something like "1couple, hetero, woman gives male a blowjob while on her knees" and then "1girl, blonde ...". It also might help to put "solo, solo focus" in the negative prompt. Depending on what you want to do it might also help to use a specific lora that is trained on showing 2 people doing whatever you want them to do.

But yea, I find it very annoying as well and my "secret" here is patience and brute *****.

Thanks for this tutorial. Really enjoy the posts

Very useful. I entered in the AI image generation world thanks to this post.

That makes me so happy! Yaye! Would you mind sharing some of your creations? I am always curious to see how other people do it and maybe I can learn something, too.

I was playing around with a new style as well, more "reduced" and black and white. What do you guys think?

And just for fun, here are some older pics I made when I started

And some size comparison

cDupin
Can you do it with this picture???

great works
.
Submissive Cuckold - lives for many years in a female-led marriage with a cuckold lifestyle.