stable-diffusion / AI - PART TWO

[Johnny Cage]

Decided to go for more of an urban pool environment to try maxhitman's more subtle wet skin prompts.

A very different question (sorry if this is beginners stuff): how do you pose your models in more complex poses?
I have tried a few things and it seems to me like the only three positions the A.I. will consistently create is standing, sitting and squatting. Anything more complex or vague than that is hard to do. e.g. 'bending over table'.

[maxhitman]

Originally Posted by Johnny Cage

Decided to go for more of an urban pool environment to try maxhitman's more subtle wet skin prompts.

A very different question (sorry if this is beginners stuff): how do you pose your models in more complex poses?
I have tried a few things and it seems to me like the only three positions the A.I. will consistently create is standing, sitting and squatting. Anything more complex or vague than that is hard to do. e.g. 'bending over table'.

Cool image once again. Congrats!

You could try using (parenthesis) to make that particular word or script
have more importance for the A.I. not to miss it, but in a certain way it
also depends on what "checkpoint" you are using as your "MAIN" checkpoint.


There are also 3 other checkpoints/safetensors in the A, B,and C sections
of the Menu which are just "extras", which usually work with your Main
Checkpoint to help out with more information and data.
According from what I have read, If your main checkpoint was trained
previously on using such "body positions", then it can very well create them
in your images, right from the start.
Now there are people making "Textual-Inversions" and "LoRA" which already
come prepared to do these things.

If your AI was not trained previously on doing these things , then your A.I.
may have a more dificult time to gather enough data to create those images.

The A.I. WILL CREATE anything you ask of it, including body positions,
but it needs more info and data to create it. It needs to be "TRAINED" by
YOU in your own computer and creating MORE Images, until your
personal A.I. is "trained" and has "learned" how to create such
things. If you do not train it in making images, it will take you
longer to get what you want.

As an example, my personal A.I. has been working for me, every day,
since January, so by now it´s so well trained , it can create most anything
I ask of it within 8 to 10 images rendered. Some times it does it right
from the start!
When I first began using SD it took a while until the A.I. began to
create exactly what I wanted to see. Some times I had to wait until the
AI created at least 25 to 30 images. Of course, back in January, I did
not know that much about writing a better Prompt for the A.I. to understand
things better. Now I do, and I am coming up with some wild and
very interesting discoveries and Prompts!

Getting back to the "checkpoints" thing...
There are numerous Photo-Realistic Checkpoints (a.k.a. Safetensors) which
are prepared for all types of body positions, including hardcore sex.
So, using one of these will surely create the images faster for you.
But you still need to write a better Prompt asking exactly what you want
yur personal A.I. to create.

You may have to download and use one of these Checkpoints, to create what
you want in a faster way, instead of training your personal A.I. numerous
times until it can create those body positions.
Search in Civitai -there is one for every taste and art-style anyone
could ask for.

The most used checkpoints which offer these types of body positions and
other sexual positions are the ones created for ANIME Art. These are Art
styles and the A.I. can easily render the human body positions, hands and
feet quite well, compared to if it was rendering realistic looking
photographs.

Let me explain,
the reason you do not often see hands and feet in Photo
realistic A.I. created images, is because this A.I. is still in its infancy
and has not mastered yet the way to better create realistic human hands,
fingers, and feet. However, it can create them quite well IF the image
is in a drawing "art style", such as Japanese ANIME and Hentai styles.

Since this S.D. is evolving SO Fast, I can very well see it creating more
realistic hands, fingers and feet withing the next 6 months or
at least by the end of this year.
So, the best is yet to come !

Some people have discovered a way to MIX some checkpoints and thus create
a whole new selection of information and data which this A.I. can use to
create better looking renderings of hands/fingers/feet for a
photo-realistic image.
But it is not a totally "photo-Realistic" image. It is a style of art which
some of us artists call 2.5-D art. It is not a 2-D flat-drawing artwork,
and it is not a 3-D image render.
It is a mix of both.
This method makes it possible for the A.I. to render a better human figure
in many body parts, but the final image will never look like a realistic
photo.
When you visit Civitai, click on the "Highest Rated" and you will see
some of these 2.5-D style checkpoints. For examplke "Perfect World".
Try them out to see the way the A.I. creates images.


You have to take things STEP-BY-STEP ... like, Baby steps.
Go slowly and learn from your mistakes. Write down on a NOTEPAD what are
the "things" and "Prompts" which works the best for you.
That is the only way to learn how to work with this new A.I.

[SynchroDub]

I saw, already, that someone started to make questions over things that are not related to SD.
People, avoid derail this thread with questions that are not linked to SD, please.
If you wanna know how AI works, Google some articles, get your facts somewhere else.
This thread is only for questions regarding Stable Diffusion, not the AI technology.
Keep it that way, please.

Thank you.

[maxhitman]

-

-

-

The old version 1.5 (last released in Dec. 2022) you are able to
create Photos & Art with NSFW/Nude content.

The newer version 2.0 has this ability completely censored. Which means,
it does not do "nude" images or NSFW (Not Safe For Work) images.

Both versions are able to create superb photo-quality images or artworks in
ANY Style. Not much diference between them. Both are great.

I have read that the version 2.0 is able to create better looking hands and
fingers, but from the images created/posted so far, I have not seen much
diference. Both versions still struggle a little with the A.I. in creating
proper hands, fingers, feet and toes. If the models wear shoes or gloves,
this problem will not be noticeable too much.
However,
since everyone wants to create "Nudie" images, the Version 1.5 is the version
which has the most special Add-ons made for it.
Check out Civitai for daily updates and for your download needs.

---------------------------------------------------------------------------------
---------------------------------------------------------------------------------
Best Samplers: (most used by many people)

Euler a
DPM++ 2M Karras
DPM++ SDE Karras
PLMS
---------------------------------------------------------------------------------
---------------------------------------------------------------------------------

CFG scale: The CFG scale adjusts how much the image looks closer to the prompt and/ or input image.
If CFG Scale is greater, the output will be more in line with the input prompt and/or input image,
but it will be distorted. On the other hand, the lower the CFG Scale value, the more likely it is
to drift away from the prompt or the input image, but the better quality.

(most used by many people)

Portraits: 7-9
Landscape: 9-12

Steps:
Portraits: 30-40
Landscape: 40-100
---------------------------------------------------------------------------------
---------------------------------------------------------------------------------

SETTING UP IMAGE SIZES

By nature, your version 1.5 will only create images in the size of 512x512,
but you can write it to make images 512x768. It can make even bigger images
but the quality will not be so good.

Recommended Resolution:

2:3 ratio (i.e 512x768)
1:1 ratio (i.e 512x512 or 576x576)

Create your images in these sizes and later UPSCALE them using
Your In-Built SD Upscaling program. Everything is in there to do it.
If you wish, you can also use free Online websites or some other
Home program you can download from the internet.
Such as Topaz Gigapixel A.I.

---------------------------------------------------------------------------------
---------------------------------------------------------------------------------

Useful Links and Video Tutorials

............................................................

HOW TO INSTALL S.D.

Stable Diffusion 1.5 - Windows Installation Guide [Tutorial]
- Check the links below video for ALL the necessary downloads 
you will need to install SD into your Home Computer.
Follow his instructions, word for word, and you will have your
won personal A.I. Stable-Diffusion working.

http://youtube.com/watch?v=ycQJDJ-qNI8

............................................................

Stable Diffusion Tools  
http://sdtools.org

S.D. Manual 
http://sdtools.org

S.D. on Reddit  
http://****************/r/StableDiffusion/

............................................................


HOW TO MASTER AI UPSCALING + UP-RES Your Images -- Stable Diffusion Web UI
http://youtube.com/watch?v=fjWcRU7Ue-0

.............................................................

How to Create "LORA"
LORA: Install Guide and Super-High Quality Training - with Community Models!
http://youtube.com/watch?v=9MT1n97ITaE

.............................................................

ULTIMATE FREE LORA Training In Stable Diffusion! ---> Less Than 7GB VRAM!
http://youtube.com/watch?v=70H03cv57-o

.............................................................

How to Create Textual-Inversions (also known as Embeddings)
ULTIMATE FREE TEXTUAL INVERSION In Stable Diffusion! Your FACE INSIDE ALL MODELS
http://youtube.com/watch?v=2ityl_dNRNw

.............................................................

AMAZING NEW Image 2 Image Option In Stable Diffusion! Apply Styles In Seconds
http://youtube.com/watch?v=VdDhGjkKD1A

.............................................................

stable diffusion how to use img to img
http://youtube.com/watch?v=c8TGgKrr7s8

.............................................................

- How to UPSCALE images using Stable-Diffusion
http://***************************/ai-upscaler/

.............................................................

- Useful websites to get ideas for writing prompts

Lexica 
http://lexica.art

Instant Art 
http://instantart.io

Prompt Hero 
http://prompthero.com

Danbooru (Anime style)
http://danbooru.donmai.us

Endless Characters 
http://endlesscharacters.com/index.php?/most_visited


.............................................................

Stable Diffusion Textual Inversion Concepts Library 
· Browse through objects and styles taught by the community to 
Stable Diffusion and use them in your prompts.
http://huggingface.co/sd-concepts-library
.............................................................

Top 500 Artists Represented in Stable Diffusion 
http://urania.ai/top-sd-artists
.............................................................

SAVE THESE VIDEOS for Yourself !! ... Top Tutorials !

Fixing Eyes With Img2Img Inpainting In Stable Diffusion ! (in 1 minute)
http://youtube.com/watch?v=EKRQb6Lc6tU

Stable Diffusion 1.5 Inpainting is AMAZING
http://youtube.com/watch?v=rYCIDGBYYnU

Inpainting in Stable diffusion for beginners
http://youtube.com/watch?v=92nTLzIXIJ0&t=9s

Inpaint whole picture, only masked, and padding explained for Automatic1111
http://youtube.com/watch?v=1xAo7ZCmyzE


HOW TO MASTER AI UPSCALING/UP-RES | Stable Diffusion Web UI
http://youtube.com/watch?v=fjWcRU7Ue-0

.............................................................

[Johnny Cage]

Thanks for the help again, maxhitman. So the problem with my lack of decent results for the bent over position is because the model I am using has not been trained enough in it.

So I tried again.
Prompt for a sexy girl in a dress bending over a pool table.

It took a while before I got decent results.

This was after like just 3 iterations:



Pretty crappy all around. Wrong position, model looks totally unnatural.

Then around the twelfth time I ran the prompt, this is what I got:


Still not great, but the stance is already vaguely like what I had in mind. Not exactly, but getting closer.

This was dozens of iterations later:

The hands are not great, the pool table looks janky, but the model is standing more or less in the way I was trying to achieve!

[maxhitman]

Originally Posted by Johnny Cage

Thanks for the help again, maxhitman. So the problem with my lack of decent results for the bent over position is because the model I am using has not been trained enough in it.

So I tried again.
Prompt for a sexy girl in a dress bending over a pool table.

It took a while before I got decent results.

This was after like just 3 iterations:

.....//......

Pretty crappy all around. Wrong position, model looks totally unnatural.

Then around the twelfth time I ran the prompt, this is what I got:

.....//......

Still not great, but the stance is already vaguely like what I had in mind. Not exactly, but getting closer.

This was dozens of iterations later:

.....//......
The hands are not great, the pool table looks janky, but the model is standing more or less in the way I was trying to achieve!

Welcome back!
Are you the same Johnny Cage who just began posting some Goodies
over at Civitai ?
If you are, welcome to Planetsuzy.
I will test out your cool stuff and see if I can "break it"
I´m just joking, but I do love to push my A.I. to its limits and
see what crazy things it can create for me. So far, this A.I. is not even
getting tired ! LOL It keeps surprising me all the time.

Those pictures look pretty decent, but don´t expect too much from the
A.I. since its not been trained to do such positions from the start.
Do NOT WORRY, it will soon begin doing them.

It does take some patience on your side and for YOU to create as
many images of the "same theme" over and over for a while,
until your personal AI can create them perfectly.

I would say that after about 25 to 40 images created it will start making
some amazing things on the "theme" you really want to see.
That is what I do when I am creating Celebrity Faces, without using any
LORA or Textual-Inversions.
From the images which were created, there are always some that
you can save for something later on. The rest, just erase them.
This AI can create a billion images and none of them will ever look the
same, so you will always have tons of images to save later.

You could always try using someone else´s image "Seeds" and try to
re-create the same image, but your pictures will ALWAYS come
out diferent. No matter what, This A.I. does not do copies perfectly.
Each picture is unique.

So , don´t lose your nerve and keep calm. Just do another batch of
six images if the previous ones did not come out too good.
This is always great training for the A.I. , so it can learn better how to
create such things you ask of it.

As we speak, I am ordering my A.I. to create for me a realistic
image of cute Naomi Watts next to King Kong on Skull Island.
I am guessing my A.I. does not know who "King Kong" is, or what is a
"Skull Island", so I am testing all sorts of prompt scripts and words
until I can find a "special Trigger word".
Once the A.I. can understand the meaning and create some thing from
that "trigger word" , I can then use it in future Prompts andmy AI
will always create what I want to see

YOU take Notes of all the useful prompts you are using in your
Positive Prompts, so you can use them in the future, for making your
model girls "standing and bending over a table, with a seductive pose"

[pelham456]

pix 2 and 3 are great. pic 1 the only problem is head looks UNATTACHED!!

maxhitman, can i ask about use of PARENS in prompts?

MOD EDIT
off-topic comment deleted per section rules

[maxhitman]

For using (parenthesis) in your prompts, and other useful information,
here is a guide...

Stable Diffusion prompt: a definitive guide

Updated March 3, 2023 By Andrew
Categorized as Tutorial - beginner, text-to-image

http://***************************/prompt-guide/


......................................................

How to Write an Awesome Stable Diffusion Prompt
by NICK LEWIS

http://howtogeek.com/833169/how-to-write-an-awesome-stable-diffusion-prompt/

......................................................

Stable Diffusion BEST Tutorial for Prompts, Beautiful Results 
- Master Prompts for Stylized Art - by AI ART ALCHEMY
( 1 hour in--depth tutorial video )

http://youtube.com/watch?v=9ErPUGlifQQ

......................................................

ON-LINE BOOK

http://openart.ai/promptbook

[Johnny Cage]

Originally Posted by maxhitman

Welcome back!
Are you the same Johnny Cage who just began posting some Goodies
over at Civitai ?

I am not! Strange coincidence.

[Johnny Cage]

Experimenting a bit with merged checkpoints.

This is one I made (of a pretty naked girl who evidently has been a bit naughty in a pedestrian tunnel) with a merge of Deliberate and URPM.



I don't know if this is because URPM has a more extensive poses keyword collection but the merged checkpoint seems to make it a lot easier to more consistently pose the girls in just the right way.