stable-diffusion / AI - PART TWO

[BoInk2]

Nice work, maxhitman!

Have to figure out what to do about the fucked up fingers
and somewhat weird nipples and belly buttons though.
Any idea?

[maxhitman]

Originally Posted by BoInk2

[center]Nice work, maxhitman!

Have to figure out what to do about the fucked up fingers
and somewhat weird nipples and belly buttons though.
Any idea?

OOOps... I forgot to mention the "trigger word" for this LORA of Anya Olsen...

The special "Trigger words" are...
a anyaolsen woman

Sorry guys and gals,
When using any LORA, Checkpoints, Textual-Inversions, you must always
use and write into your prompts any and all "trigger words" which will help
your AI to better create the image better.

I will also post here below all the trigger words I used in the BLIP image captions
to train the AI to better create her face. Notice there are 3 images with her
doing "fellatio", which I included in the training set and which you can create
such images if you which.
This way the AI will also create her facial expressions much better
if you are inclined to create a more realistic Porn-style image.
Personally, I intent to use her as my Artistic-Nude-Photography Model and nothing
to do with Porn. Anya has a beautiful body and bright blue eyes which are
just perfect for natural nude Photoshoots such as the ones often seen in Museum
quality photographs.


"an1a0lsen_v1.0 a anyaolsen woman with a white lace bra is posing for a picture": 1,
"an1a0lsen_v1.0 a anyaolsen woman with a smile on her face": 1,
"an1a0lsen_v1.0 a anyaolsen woman is posing for a picture outdoors": 1,
"an1a0lsen_v1.0 a anyaolsen woman in a bra top sitting on a chair": 1,
"an1a0lsen_v1.0 a anyaolsen woman with a small breast is posing naked": 2,
"an1a0lsen_v1.0 a anyaolsen woman in a lingerie posing for a picture": 1,
"an1a0lsen_v1.0 a anyaolsen woman with a shirt is smiling": 1,
"an1a0lsen_v1.0 a anyaolsen woman in a bra top and panties is smiling": 1,
"an1a0lsen_v1.0 a anyaolsen woman with a pink top is smiling": 1,
"an1a0lsen_v1.0 a anyaolsen woman with face make-up in a white top and a pink skirt": 1,
"an1a0lsen_v1.0 a anyaolsen woman is posing nude in front of a white curtain": 1,
"an1a0lsen_v1.0 a anyaolsen woman with a small breast is posing for a picture": 1,
"an1a0lsen_v1.0 a anyaolsen woman with face make-up in a white top and a skirt is smiling": 1,
"an1a0lsen_v1.0 a anyaolsen woman is standing in front of a window": 1,
"with a lamp": 1,
"with a smile on her face": 1,
"an1a0lsen_v1.0 a anyaolsen woman in a lace bra top with a smile on her face": 1,
"an1a0lsen_v1.0 a anyaolsen woman in a lace lingerie is posing for a picture": 2,
"an1a0lsen_v1.0 a anyaolsen woman is standing with a smile on her face": 1,
"an1a0lsen_v1.0 a anyaolsen woman in a lace bra top and panties": 1,
"an1a0lsen_v1.0 a anyaolsen woman in a white bra top sitting on a couch": 1,
"an1a0lsen_v1.0 a anyaolsen woman in a white lingerie posing for a picture": 1,
"an1a0lsen_v1.0 a anyaolsen woman is smiling for a picture": 1,
"an1a0lsen_v1.0 a anyaolsen woman is smiling and posing for a picture": 1,
"an1a0lsen_v1.0 a anyaolsen woman is sitting on a bed in a white top": 1,
"an1a0lsen_v1.0 a anyaolsen woman in a dress is posing for a picture": 1,
"an1a0lsen_v1.0 a anyaolsen woman with long hair is posing for a picture": 1,
"an1a0lsen_v1.0 a anyaolsen woman with a shirt and a smile on her face": 1,
"an1a0lsen_v1.0 a anyaolsen woman with a small breast posing for a picture": 1,
"an1a0lsen_v1.0 a anyaolsen woman in a lace lingerie is sitting on a bed": 1,
"an1a0lsen_v1.0 a anyaolsen woman in a lace bra top and long hair": 1,
"an1a0lsen_v1.0 a anyaolsen woman in a white top and pink skirt standing on a stair": 1,
"an1a0lsen_v1.0 a anyaolsen woman standing in a room with a chair": 1,
"an1a0lsen_v1.0 a anyaolsen woman in a lace bra top sitting on a bed": 1,
"an1a0lsen_v1.0 a anyaolsen woman with long hair is posing naked": 1,
"an1a0lsen_v1.0 a anyaolsen woman in a black bra top": 1,
"an1a0lsen_v1.0 a anyaolsen woman with a dress is sitting on a couch": 1,
"an1a0lsen_v1.0 a anyaolsen woman with a small breast is making a orgasm face": 1,
"an1a0lsen_v1.0 a anyaolsen woman in a room with a lamp is posing for the camera": 1,
"an1a0lsen_v1.0 a anyaolsen woman in a white dress is smiling posing for a picture": 1,
"an1a0lsen_v1.0 a anyaolsen woman with a sweater is smiling": 1,
"an1a0lsen_v1.0 a anyaolsen woman with long hair is smiling": 1,
"an1a0lsen_v1.0 a anyaolsen woman in a pink top and jeans shorts": 1,
"an1a0lsen_v1.0 a anyaolsen woman in a lace bra top is sucking a man's penis cock": 1,
"an1a0lsen_v1.0 a anyaolsen woman in a bra top is licking a man's penis cock": 1,
"an1a0lsen_v1.0 a anyaolsen woman in a lingerie is licking a man's penis cock": 1,
"an1a0lsen_v1.0 a anyaolsen woman is making a orgasm face expression for a picture": 1,
"an1a0lsen_v1.0 a anyaolsen woman with face makeup is posing for a picture": 1,
"an1a0lsen_v1.0 a anyaolsen woman in a garterbelt and stockings is sitting on a couch": 1,
"an1a0lsen_v1.0 a anyaolsen woman with a smile is posing in front of a white wall": 1
.............................................................

Those are some really cool and sexy AI-pictures, BoInk2 man
Thanks for sharing ! I love them all.
Anya is such a cute and wonderful woman.
My type of woman in bed too ! I´m also a nymphomaniac

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

Below are some of my suggestions for a really nice way to create a
realistic-AI-photo.

1 - you have to be very specific in your writing of the positive Prompts.
Do not worry about how long it is, because on the Main menu there is
a little box which will tell you how long the script is and what is the maximum
number of "tokens" you can use. Be very specific as to what exactly you
want to see and how much you want YOUR A.I: to create them.

2 - Describe the image you want to see as if you were talking to someone
who is blind, or is perhaps talking to you on the other side of the telephone
in a phone conversation and is not seeing anything you are seeing.
Start by describing in detail... the head, hair color or hair style, face,
eyes, lips...
then describe the body, body type, breast size (if she is nude),
what sort of clothing does the model have dressed...
then describe in detail what type of photograph or Art type do you want to see, for example a realistic photo, or some style of painting style in particular...
Then describe the surroundings of your model, for example, what location, where ?, what is the background and what does it look like....
then describe to me once again, other fine little details your image should have in it.

Here is an example of one of my NSFW prompts...

<lora:an1a0lsen_v1.0:0.8>, a Realistic Full Portrait of a anyaolsen woman with blue eyes and dark long hair style, Beautiful, Detailed Face, Perfect Eyes, (Eye makeup), smiling face showing teeth, (highly detailed skin:1.1), (perfect breasts:1.1), (perfect nipples:1.1), perfect slim body, NSFW, Nude, wearing a white lace lingerie with stockings, Modelshoot style, Professional Photography, Soft Lighting, Photorealistic, detailed textures, standing in a living room background, blurred background, volumetric fog, RAW, analog style, Erotic, sharp focus, 8k, HD, DSLR, high quality, Fujifilm XT3, film grain, award winning, masterpiece

By describing with as much detail as possible what you would like the AI to
create for you, the AI will pay attention more closely to your written
prompt and will make it. If you write your description as I suggested
above, in describing the image in detail by "sections" or by "Parts", the
AI will understand better what you want to see.

To recap what I wrote before, these would be the "Sections" (or parts)
in an image you should describe...
1 - Head part (hair style / hair color / detailed face or not? / hat? /other)
2- Face part (detailed eyes? / lipstick ? / smiling or sad? /other)
3 - Body part ( breast size? / slim? /Athletic ? / tall ? /Slim? / etc.)
4 - Include here the "trigger words" for the checkpoints you are using
5 - describe what the model should be wearing or what is she/he doing
6 - describe what type of lighting techniques in your image
7 -describe the background where the model is located in
8 - include extra information to make sure the AI will create them
into your image
9 - describe it AGAIN to the AI that you want this to be a "award Winning"
image, or a "masterpiece" or a "curated collection" or a Museum quality"
type of image

If you start seeing alot of problems with "nipples" and "bellybuttons" or
whatever body part,
start writing in your POSITIVE PROMPT...
(perfect nipples:1.1) and (perfect body:1.1)

ALSO write into your NEGATIVE PROMPT...
Malformed nipples, deformed nipples, deformed body, deformed belly button,

By doing this, your AI will try better to create a perfect body.


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

As for the "Hands and fingers" problem, this AI is now beginning to create
them more precise and better.
This also depends on the Checkpoints you use to
help create the image and ALSO on your Prompts, AND also how much you
set your weights (number) in the CFG Scale when creating images.
If you set your CFG-Scale to around 5 or 6, the hands and finger's are better
created by the AI, but the quality of image will look like it was "painted" and
not a full realitic-type of photograph.

You can also include into your prompts the words...
(perfect hands and fingers:1.1)
which will help out alot , IF you start seeing many
images of hands being created by the AI.
However, this AI does try to conceal hands from being in view and
usually creates images with hands being hidden.

By using "ControlNet", this problem will soon be gone.
There are ways that the ControlNet program is making sure the AI is
creating better hands and fingers.

For better hands I suggest this LORA...

EnvyBetterHands -LORA from Civitai
http://civitai.com/models/47085/envybetterhands-locon

900 Hands Library for Depth Library - ControlNet -LORA from Civitai
http://civitai.com/models/67174/900-hands-library-for-depth-library-or-controlnet

50 Hands Images for Depth Library - ControlNet -LORA from Civitai
http://civitai.com/models/67202/50-hands-images-for-depth-library-or-controlnet

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

ControlNet Helps You Make Perfect Hands With Stable Diffusion 1.5
http://mpost.io/controlnet-helps-you-make-perfect-hands-with-stable-diffusion-1-5/

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

Video tutorials and other ideas...

Stable Diffusion ControlNet Depth Library Extension -Easy hands - Controlnet Guidance Start Hands
http://youtube.com/watch?v=I9dva9EBhl8


ControlNet Guidance tutorial. Fixing hands?
http://youtube.com/watch?v=wNOzW1N_Fxw


GET PERFECT HANDS With MULTI-CONTROLNET & 3D BLENDER! This Is INSANE!
http://youtube.com/watch?v=ptEZQrKgHAg

As for the video above  tutorial, another way I can suggest to create better 
hands taken from another image, is to use a photo of a hand position
 downloaded from a picture as posted in Deviantart (lots of models there), 
or from VirtuaGirl nudei-photos website, or even from photographs you 
see here on Planetsusy models. 
The resources to find that perfect hand position are many.

I hope this help out

I find less in a prompt is better but be more focusing on the prompt word, you can fix eyes and other bits in img2img

You Tube

You Tube

[maxhitman]

Someone asked me:
" Why my downloaded LORA files have no Preview picture on them ?"

When you have your Stable-Diffusion program Menu opened and you go
to your Model directory to look for a "Lora", or "Textual Inversion" or "Checkpoint",
you usually just see a Grey square.
Some times this may get confusing to search for what you want, because you may
have 25 or more models in there and some may have strange names written
on them.

YOU must ADD into the same folder (where that downloaded model is located at)
an image ....in PNG or Jpeg... rename it with the same name as your Model.
Once the image is inside that folder, hover your mouse-pointer over the grey-square-box
and you will see a "replace preview" sign appear. Click on it and automatically the
Model will now have an image on it.
- If you erase the image, the model will once again go back to a grey-square.

Once you have all your models with images on them, it will be easier to pick them out
to work with (specially if they have a confusing name on them).

At the moment I have not found a way to include into the training-Lora any images
to embed them permanently into them.

Are your AI Previews sexy ? Try this EASY FIX!

You Tube

You Tube

[BoInk2]

Thanks maxhitman, I'll keep trying. Although I used e.g. perfect body, perfect breasts, perfect hands, etc. already. I'll check if that BetterHands Lora helps.

And to all Stable Diffusion guys, nVidia released a new driver with this in its change log:
"(...) Additionally, this Game Ready Driver introduces significant performance optimizations to deliver up to 2x inference performance on popular AI models and applications such as Stable Diffusion."

[BoInk2]

Interesting to see the different results per model used:

1st uberRealisticPornMerge_urpmv13:



2nd endlessreality_v11:



Fingers still not much better, despite BetterHands Lora.

3rd realisticVisionV20_v13:



Unfortunately realisticVisionV20_v20 seems to deliver weird nipples,
and additional nipples all the time, doesn't matter the negative prompt.
Even v13 gives some weird things on the nipples, like the dark dot in the pic above.

[BoInk2]

One more with a new model called

epicrealism_newEra:



Nipples still not perfect, seems her hair got in the way.

And one Kristen with that model:

Gotta be honest RealisticVision is the best IMHO

[maxhitman]

Originally Posted by BoInk2

Unfortunately realisticVisionV20_v20 seems to deliver weird nipples,
and additional nipples all the time, doesn't matter the negative prompt.
Even v13 gives some weird things on the nipples, like the dark dot in the pic above.

Great photos! Thanks for sharing and also testing the different Checkpoints.

To make SURE the A.I. creates perfect breasts and Nipples, make sure to
ALWAYS include in your prompts the following words...
This was from my previous posted NSFW Prompt

<lora:an1a0lsen_v1.0:0.8>, a Realistic Full Portrait of a anyaolsen woman with blue eyes and dark long hair style, Beautiful, Detailed Face, Perfect Eyes, (Eye makeup), smiling face showing teeth, (highly detailed skin:1.1), (perfect breasts:1.1), (perfect nipples:1.1), perfect slim body, NSFW, Nude, wearing a white lace lingerie with stockings, Modelshoot style, Professional Photography, Soft Lighting, Photorealistic, detailed textures, standing in a living room background, blurred background, volumetric fog, RAW, analog style, Erotic, sharp focus, 8k, HD, DSLR, high quality, Fujifilm XT3, film grain, award winning, masterpiece

By placing the words inside a bracket-Paranthesis you will ORDER the A.I.
to pay more attention to these words and create them as perfect as possible.
(highly detailed skin:1.1), (perfect breasts:1.1), (perfect nipples:1.1), perfect slim body

By adding to the words a set of numbers such as
(perfect breasts:1.1)
or (perfect breasts:1.2) or (perfect breasts:1.3)
This will make these words even more important, so the A.I. will give them
more weight and importance in your written script.


There is also a proper way you can write a NEGATIVE Prompt which is
just as important as your positive prompt on the things you want to see
being created.
Posted below is my most used negative Prompt.
The things I DO NOT want to see the A.I. create in my images...

This would be for a NSFW image...

((Clothes)), (pants), (bikini), ((bra)), (underwear), (dress), canvas frame,((ugly eyes, deformed iris, deformed pupils, deformed lips and teeth:1.1)), un-detailed skin, semi-realistic, cgi, 3d, render, sketch, cartoon, drawing, anime, text, monochromatic, cropped, out of frame body, worst quality, low quality, jpeg artifacts, ugly, duplicate, morbid, mutilated, extra fingers, mutated hands, poorly drawn hands, poorly drawn face, mutation, deformed, blurry, dehydrated, bad anatomy, bad proportions, extra limbs, cloned face, disfigured, gross proportions, malformed limbs, missing arms, missing legs, extra arms, extra legs, fused fingers, too many fingers, long neck, doll, child, un-detailed skin, monochrome, black and white,

for a SFW (Safe for work) image I would use this one...

(NSFW), (Nude), (Naked), (pussy), (vagina), (penis), (nipples), canvas frame,((ugly eyes, deformed iris, deformed pupils, deformed lips and teeth:1.1)), un-detailed skin, semi-realistic, cgi, 3d, render, sketch, cartoon, drawing, anime, text, monochromatic, cropped, out of frame body, worst quality, low quality, jpeg artifacts, ugly, duplicate, morbid, mutilated, extra fingers, mutated hands, poorly drawn hands, poorly drawn face, mutation, deformed, blurry, dehydrated, bad anatomy, bad proportions, extra limbs, cloned face, disfigured, gross proportions, malformed limbs, missing arms, missing legs, extra arms, extra legs, fused fingers, too many fingers, long neck, doll, child, un-detailed skin, monochrome, black and white,


If you are going to create an image with a woman dressed don´t even mention
the word Breasts or nipples! NEVER use it in your prompts.
This A.I. is a very horny machine and will create some
breasts in view in some images ! LOL
......................................................

Keep in mind something that most people tend to forget when using
Stable-Diffusion.
Whatever "Checkpoint" you use as your MAIN-Checkpoint, make sure
what types of images it is good for.

Some Checkpoints are best at creating "painted style" images, while
others tend to create more ANIME style artworks or pictures and if you try
to create a realistic style photo of a woman, your model will have a doll
like face with some asian characteristics on their faces.
Other checkpoints say they create realistic photos but they actually are a mix of ANIME-painted style art and photo-realism which kind of downgrades
the quality of the images a little and the models kind of look "painted".
Some times, these may be good for creating better looking hands and
fingers, but will not create for you a very realistic overall image.
It is up to you to decide which you like to use best and keep experimenting
until you discover a Checkpoint which does everything you would like
to see in an image. I personally have tested over 40+ Checkpoints until I decided on the ones I like the most, depending on which type of images
I like the AI to create.

For my realistic looking photoshoot Photos (nude and clothed women)
I like to use Realistic Vision, but not the new version 2.0,
the best for me is version 1.3, which is still on the Civitai webpage which
you can download and use. Also try version 1.2.
I have no idea why the creator of this great Checkpoint went to
create a version 2.0 , which kind of downgrades everything that was so
great about the version 1.3, and version 1.2 .
These older versions are super great in making tons of known celebrities
without the use of any LORA or T.I.´s, AND they contain much more information for creating amazing pictures.
This is always my go-to-first checkpoint for making photos... version 1.3

I also use version 1.3 to train the A.I. to create LORA and the final results
are way much better than any other checkpoints. The details on the faces
are impressive, more realistic, unlike other checkpoints.

If you would like to experiment with your A.I. to create realistic
looking photos, I will then suggest Realistic-Vision 1.3 as your main checkpoint. It already comes with the vae-ft-mse-840000-ema-pruned.ckpt included!

Then in your extra checkpoints merges, in the Menu Section "Checkpoint Merger" , which are A, B, and C, I use these ...

Main checkpoint - RealisticVision V.1.3

A - Protogen X34
B - Clarity_version 1.9
C - Deliberate (the first version)

By combining the information inside these other checkpoints to merge with
your main checkpoint, you get more useful information for the A.I. to
create better pictures.
For example, if your main checkpoint might be missing some information
to create a "painted style" portrait, or maybe a cyborg-robot looking woman,
then the AI will search and look into the other checkpoints (A, or B, or C)
to retrieve useful information so it can create a better image for you.

With this in mind, don´t just use ONE checkpoint as your main checkpoint.
Always include another or maybe 2 , and place them into the
spaces for the A and B checkpoint merger section.

I was using V1.3 but about a month ago I went over to V2 and have never looked back because the rendering is far superior in every way as long as you tweak the prompts.

As V1.3 was prompt heavy you had to include a lot of information for the subject matter but with V2 you can put less information in the prompt as V2 is better at learning which a lot of people just don't get.

Checkpoints are becoming better at learning as the whole AI is getting better and is now getting confused by too much information in a prompt, you have to change along with the AI, negative prompts can stay the same though.

This image I just created is a perfect example, V1.3 makes it a bit odd looking but with a prompt of less information I got a much better picture and all I asked was.

A full frame photo of Sandra Bullock 30 year old



If you are still getting odd photos then I would suggest proof reading your prompt as something might be conflicting, I can not check your png info as downloading from an upload to a image host removes the info/exif data.

Also make sure you are using a Random Seed and different Sampling Methods, and also check to see what methods and steps do what.

http://artstation.com/blogs/kaddoura/pBPo/stable-diffusion-samplers