stable-diffusion / AI - PART TWO

[maxhitman]

I am happy to say that today I have installed all the useful gadgets necessary into
the Stable-Diffusion to create Textual-Inversions (a.k.a. "Embeddings") and LoRA´s!
It was a little dificult and it took me about 2 hours! (I am not a computer wiz-kid)
I will give everyone a good and easy tutorial so you can also do it, if you
are interested in creating some unique things with the S.D.
At the moment I am testing it to create my first Lora of an actress which will be
needed in some cool action scenes involving "aliens".
Its the face of a younger (1980) Sigouney Weaver, which many people and
film critics consider her to be Hollywood´s first female action hero !
"Kill those nasty aliens Ripley ! Game Over Man... game over !! "

The 2 Lora´s already available at Civitai do not create her face too well
and just using the normal checkpoints (such as Realistic-Vision),
this A.I. still struggles to create her face
properly. Often you see her with a huge forehead and distorted face
features or eyes. I plan on fixing this problem with my own selection
of images taken from film screenshots and photoshoots, fixing them up
with Photoshop and then train this A.I. to make her face like the
real "Ellen Ripley".
So far, the first tested Lora came out very cool with a run of 3000 steps on
a selection of 30 pictures. That is like 3000 images created to teach the A.I. how
to make her face, in about 25 minutes. BUT I want more face details!
So, I am now making another LoRA with 4500 steps, and perhaps more.
I will share with you my finished Lora so you can also test it in your S.D. and
see how it works

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

I will answer a question here, which someone asked me in a PM.

Someone asks,
Are there more places to download useful LORA or Textual-Inversions for
making more "Ai Porn" ?

Don´t be embarrassed. Every new invention will either be used for
making a new weapon or be used for making Porn. We humans are
the craziest species in the galaxy !
Using the Stable-Diffusion A.I. for making "home-Porn" is no exception.
Just make sure you don´t go too crazy with it and start making bad porn !
...and do wear a condom !


A.I. fans already know about Civitai.com to get their free downloads of
all things related to S.D. (Checkpoints / LoRA/ Textual Inversions /Styles / etc.) , of which many are for creating "porn" of all types and
also "Anime Porn art".
But
there are other not well-known places where you can also find free
stuff if you look around.... Look below....

This is a place with thousands of such "Ai Porn" goodies you can
download and use with your personal A.I. at home.
You will probably spend 2 Weeks visiting this webpage every day
just to download everything they have ! Just remember to sleep and
eat some food while you´re at it.

The Ai Porn Guide

http://gitgud.io/gayshit/makesomefuckingporn#lora-list

The Ai Porn Guide webpage is not set up like Civitai with preview images
of each download.
You will have to click the Links, which will open up a folder in a remote server, usually from Mega.
Inside that folder you can look at the diferent image previews of what
that Lora, or Textual-Inversion, or Checkpoint can create for you.
If you like it,
then just download the necessary files and install into your S.D.

~~~~~~~~~~~~~~~~~~~~~~~~~~~~

The famous "Orange Mixes".
These are for ANIME style Art which have alot of naughty things
going on in the art. Some are already posted as "checkpoints" at Civitai,
but there are even MORE mixes from this creator (WarriorMama777)
and many other Ai artists.

You can find them at ...

OrangeMixs
http://huggingface.co/WarriorMama777/OrangeMixs

~~~~~~~~~~~~~~~~~~~~~~~~~~~~

From Rentry comes another HUGE List of all things related to S.D.
but you will have to check the list careful and then download what you
want or need.

http://rentry.org/sdmodels

~~~~~~~~~~~~~~~~~~~~~~~~~~~~
( IMPORTANT - Planetsuzy will not be responsible if these things
break your Ai program, so be careful and use your Anti-Virus to scan
the downloaded content before you install it on your computer )

~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Originally Posted by maxhitman

The famous "Orange Mixes".
These are for ANIME style Art which have alot of naughty things
going on in the art. Some are already posted as "checkpoints" at Civitai,
but there are even MORE mixes from this creator (WarriorMama777)
and many other Ai artists.

I would stay clear of Orange Mixes especially merged as they are being warned for very close to the knuckle age limit stuff.

[Johnny Cage]

I have tried to install the ControlNet extension for the Automatic1111 webui.
I believe I have followed the correct steps. The option does show up in the webui.
And I can select the models and everything. But when I run a prompt with a pose active it does not change the pose of the img I am creating. What could be the problem?

[Johnny Cage]

Never mind, figured it out myself, seems to be a memory problem with my GPU, if I reduce the resolution it does work.

[maxhitman]

Here is something for you to see Johnny_Cage (and everyone interested)

..about Control-Net

ControlNET Stable Diffusion: Automatic Pose Transfer - Easy Guide
http://youtube.com/watch?v=ci7NfTsifd0&t=0s

LORA with ControlNET - Get the BEST results - Complete Guide - using Jenna Ortega
http://youtube.com/watch?v=6glHekd25nE

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

...Another Useful tip on how to use Inpainting easily...
It´s about that img2img-Inpainting process to fix faces , when your A.I. creates
the wrong face, or when it creates a very bad face on your models.
The image(s) can always be fixed by using these following steps...

Quickly fix bad faces using inpaint
byBernard Maltais

http://youtube.com/watch?v=IGFE1aOFZ_4

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

How to create LORA´s
LoRA stands for "Low-Rank Adaptation"

What's unique about LoRA?
Faster training: Training a new concept with LoRA takes just a few minutes.

Smaller outputs: Trained LoRA outputs are much smaller than DreamBooth outputs.
This makes them easier to share, store, and re-use.

Multiple concepts: You can combine multiple trained concepts in a single image.
Great for training art styles, faces, bodies and anything you can imagine, all in
one small package.
It´s the reason everyone is doing more LoRA´s these days.


Here are 3 YouTube videos (below) with a very good explanation and
easy step-by-step tutorial on how to install
Kohya_ss and use it with your Stable-Diffusion to create LORA´s.

I learned more from these videos than from watching many others.
On other YouTube videos they always skip some important steps.

These are the 3 videos you must watch...

step 1 - installing the Kohya_ss for creating Lora
by Olivio Sarikas  (His video tutorials are excellent!)
LORA: Install Guide and Super-High Quality Training 
http://youtube.com/watch?v=9MT1n97ITaE&t=399s


Creating a LoRA weight using kohya_ss GUI, 
part 1: Dataset creation and preparation 
- He is making a LORA of Jennifer Aniston 
http://youtube.com/watch?v=N4_-fB62Hwk&t=148s

Creating a LoRA weight using kohya_ss GUI, 
part 2: Training the model and creating outputs
http://youtube.com/watch?v=k5imq01uvUY

I often record these YouTube videos for my own Tutorial files
for later viewing, because some times the videos get deleted
by YouTube.
So save them !
If you download and use 4K-video Downloader (its free),
http://4kdownload.com/38
you will be able to save any YouTube videos (in High-Resolution)
to your computer and watch them later, even if your internet
connection fails
It also creates MP3 music files taken from YouTube Videos and from other websites.

[Johnny Cage]

I had tried out some Lora's but no textual embeddings.

Been experimenting a bit with the Nata Lee one.

[maxhitman]

Some of my test images... My first Lora sucks I want to see better.


-

[maxhitman]

some test images


-

-

-

[MaxLord]

I think I'm going to look into creating a few of my own Loras as well. What is the best tutorial to look into for that?

[pelham456]

who is the one in post 39 pls? is that based on a real celeb or a fully-created ficticious one?

HUBBA HUBBA!

nata lee also really nice, tho not quite realistic. tbh, she's better than the real one!