Installing Stable Diffusion & Dreambooth for Windows

--Resources--
Website & Videos:
Aitrepreneur SUPER Stable diffusion 2.0 Installation - https://www.youtube.com/watch?v=vg8-NSbaWZI
Aitrepreneur DREAMBOOTH: Train Stable Diffusion With Your Images (cloud install) - https://youtu.be/7m__xadX0z0

XavierXiao Dreambooth-Stable-Diffusion - https://github.com/XavierXiao/Dreambooth-Stable-Diffusion
Gammagec Dreambooth-SD-optimized - https://github.com/gammagec/Dreambooth-SD-optimized
JoePenna Dreambooth-Stable-Diffusion - https://github.com/JoePenna/Dreambooth-Stable-Diffusion
(JoePenna's github no longer focuses on local running, but still a good resource.

Automatic1111 Stable-Diffusion-webui - https://github.com/AUTOMATIC1111/stable-diffusion-webui
Automatic1111 Guide - https://stablediffusionguides.carrd.co/#one
Spinferno Dreambooth Guide - https://www.reddit.com/r/StableDiffusion/comments/xpoexy/yet_another_dreambooth_post_how_to_train_an_image/

Hugging Face - https://huggingface.co/
Birme Bulk Image Resizing - https://www.birme.net/?target_width=512&target_height=512

Installers:
Python - https://www.python.org/downloads/release/python-3109/ (During Install "Add Python to PATH")
Anaconda - https://www.anaconda.com/
Git - https://gitforwindows.org/

Downloads:
Stable Diffusion Model Weights 1.4 - https://huggingface.co/CompVis/stable-diffusion-v-1-4-original/resolve/main/sd-v1-4.ckpt
GFPGAN Face Correction - https://github.com/TencentARC/GFPGAN/releases/download/v1.3.0/GFPGANv1.3.pth

Regularization dataset examples: man_euler, man_unsplash, person_ddim, woman_ddim (https://github.com/djbielejeski?tab=repositories)

----------------------------------------------------------------------------------------------
Pre-Requirments:
1.	Install Python (3.10.x).
	During Install CHECK "Add Python to PATH".
2.	Install Anaconda.
3.	Install Git.
4.	Restart PC.
5.	Create a folder in your user directory, I named mine "AI" (C:\Users\natemac\AI)
5b.	You will need to update "natemac" to your user name for this guide.

----------------------------------------------------------------------------------------------
Install Stable Diffusion webui:
(These same steps are also in "Aitrepreneur SUPER Stable diffusion 2.0 Installation" video and "Automatic1111 Guide")
1.	Launch "Anaconda Prompt" from the Start Menu.
2.	Go into your new folder [cd AI] (C:\Users\natemac\AI).
3.	Clone Stable-Diffusion-webui [git clone https://github.com/AUTOMATIC1111/stable-diffusion-webui.git].
4.	Download "Stable Diffusion Model Weights 1.4" (You will need a Hugging Face account).
5.	Rename the file from "sd-v1-4.ckpt" to "model.ckpt".
6.	Move the "model.ckpt" file to the Stable Diffusion "Models" folder (C:\Users\natemac\AI\stable-diffusion-webui\models).
7.	Download the GFPGAN File.
8.	Move the "GFPGANv1.3.pth" file to root Stable Diffusion webui folder (C:\Users\natemac\AI\stable-diffusion-webui)
9.	Stable Diffusion webui is installed.

----------------------------------------------------------------------------------------------
Running Stable Diffusion webui:
1.	Run/Double-click on the "webui-user.bat" inside the "stable-diffusion-webui" folder (this will take longer the first time).
1b.	This is complete once you see the line "To create a public link, set `share=True` in `launch()`." in the Command Prompt.
2.	Open a web browser and go-to: http://127.0.0.1:7860
3.	Stable Diffusion is running.

----------------------------------------------------------------------------------------------
Closing Stable Diffusion webui:
1.	Close the web browser.
2.	Close the Command Prompt that opened after running "webui-user.bat".

----------------------------------------------------------------------------------------------
Install Dreambooth-SD-optimized:
(These steps are similar to "Aitrepreneur DREAMBOOTH" video)
1.	Launch "Anaconda Prompt" from the Start Menu.
2.	Go into your new folder [cd AI] (C:\Users\natemac\AI).
3.	Clone Dreambooth-SD-optimized [git clone https://github.com/gammagec/Dreambooth-SD-optimized.git].
3b.	XavierXiao's version has the same steps, but a different folder name [git clone https://github.com/XavierXiao/Dreambooth-Stable-Diffusion.git].

----------------------------------------------------------------------------------------------
Install Dreambooth-SD-Optimized - Build Environment:
1.	Open the "environment.yaml" file (in an editor such as notepad) in the Dreambooth-SD-optimized folder (C:\Users\natemac\AI\Dreambooth-SD-optimized).
2.	Edit the first line (name: ldm) to your choosing "name: SD-Optimized". Save & close.
3.	Launch "Anaconda Prompt" from the Start Menu.
4.	Go into your Dreambooth-SD-optimized folder [cd C:\Users\natemac\AI\Dreambooth-SD-optimized].
5.	Build environment [conda env create -f environment.yaml]
5b. To activate environment use [conda activate SD-Optimized]
5c. To deactivate an active environment use [conda deactivate]

----------------------------------------------------------------------------------------------
Install Dreambooth-SD-optimized - Copy Files:
1.	Copy "model.ckpt" from (C:\Users\natemac\AI\stable-diffusion-webui\models) to (C:\Users\natemac\AI\Dreambooth-SD-optimized)
2.	Activate the environment by typing into the Anaconda Prompt [Conda activate SD-Optimized]
3.	Dreambooth-SD-optimized is now installed.

----------------------------------------------------------------------------------------------
Pre-Training - Regularization Images:
1.	Create folder structure in the "Dreambooth-SD-optimized" root-folder (Dreambooth-SD-optimized/outputs/txt2img-samples/samples/) [md outputs\txt2img-samples\samples]
2.	Go-to this folder in the Anaconda Prompt [cd C:\Users\natemac\AI\Dreambooth-SD-optimized\outputs\txt2img-samples\samples]
3.	Choose one of the datasets (man_euler, man_unsplash, person_ddim, woman_ddim)
4.	Replacing DATASET with your choice (https://github.com/djbielejeski/Stable-Diffusion-Regularization-Images-DATASET.git)
5.	Run your dataset clone [git clone https://github.com/djbielejeski/Stable-Diffusion-Regularization-Images-woman_ddim.git]
6.	Open your new DATASET folder(Stable-Diffusion-Regularization-Images-woman_ddim).
7.	Copy your DATASET folder (woman_ddim) to the (samples) folder.
8.	Delete empty download folder (Stable-Diffusion-Regularization-Images-woman_ddim).
9.	You should now have 1500 images in the folder (C:\Users\natemac\AI\Dreambooth-SD-optimized\outputs\txt2img-samples\samples\woman_ddim).

----------------------------------------------------------------------------------------------
Pre-Training - Prepping Training Samples:
1.	Collect your sample images you'd like to use. At least 10.
1b.	***THE COLLECTION ON IMAGE MUST BE AN EVEN AMOUNT.***
2.	Crop images as needed to not have additional info such as another person if you are training it on a single person.
3.	Use (https://www.birme.net/?target_width=512&target_height=512) to Bulk crop and size images. Download the zip.
4.	Create a folder in the root (Dreambooth-SD-optimized) folder named "training_samples".
5.	Create a folder in the (training_samples) of your training set ("C:\Users\natemac\AI\Dreambooth-SD-optimized\training-samples\rhaenyra").
6.	Copy your finalized Cropped & Sized images into this folder.
7.	Open the "personalized.py file (in an editor such as notepad) located in ("C:\Users\natemac\AI\Dreambooth-SD-optimized\ldm\data\personalized.py").
8.	We will be editing and saving line 11 ('photo of a sks {}',) to your training name ('rhaenyra {}',).

----------------------------------------------------------------------------------------------
Training a Model with your Samples:
1.	Launch "Anaconda Prompt" from the Start Menu.
2.	Go into your Dreambooth-SD-optimized root folder [cd C:\Users\natemac\AI\Dreambooth-SD-optimized].
3.	Take note of your TRAINING-SAMPLES-NAME (rhaenyra) your REGULARIZATION-IMAGES-NAME (woman_ddim) & CLASS (woman).
4.	We will be editing the below code:
python main.py --base configs/stable-diffusion/v1-finetune_unfrozen.yaml -t --actual_resume model.ckpt --reg_data_root outputs\txt2img-samples\samples\REGULARIZATION-IMAGES-NAME -n TRAINING-SAMPLES-NAME --gpus 0, --data_root training_samples\TRAINING-SAMPLES-NAME --batch_size 2020 --class_word CLASS

5.	What it should look like:
python main.py --base configs/stable-diffusion/v1-finetune_unfrozen.yaml -t --actual_resume model.ckpt --reg_data_root outputs\txt2img-samples\samples\woman_ddim -n rhaenyra --gpus 0, --data_root training_samples\rhaenyra --batch_size 2020 --class_word woman

6. Training is complete even if you see an error as long as somewhere you see the line "Another one bites the dust..."

----------------------------------------------------------------------------------------------
Pruning & Transfering Samples Model:
0.	***If you are running Dreambooth-SD-optimized, you will need to add "prune_ckpt.py" from "XavierXiao Dreambooth-Stable-Diffusion" clone to the "Dreambooth-SD-optimized" root folder.***
1.	Find the name of the training data folder (C:\Users\natemac\AI\Dreambooth-SD-optimized\logs\) "rhaenyra2022-09-28T11-46-55_rhaenyra".
2.	Launch "Anaconda Prompt" from the Start Menu.
3.	Go into your Dreambooth-SD-optimized root folder [cd C:\Users\natemac\AI\Dreambooth-SD-optimized].
4.	Run the following command with TRAINED-DATA-FOLDER replaced:
python "prune_ckpt.py" --ckpt "C:\Users\natemac\AI\Dreambooth-SD-optimized\logs\TRAINED-DATA-FOLDER\checkpoints\last.ckpt"

5.	What it should look like:
python "prune_ckpt.py" --ckpt "C:\Users\natemac\AI\Dreambooth-SD-optimized\logs\rhaenyra2022-09-28T11-46-55_rhaenyra\checkpoints\last.ckpt"

6.	There is now a roughly 2GB file named "last-pruned.ckpt" inside the "checkpoints folder (C:\Users\natemac\AI\Dreambooth-SD-optimized\logs\rhaenyra2022-09-28T11-46-55_rhaenyra\checkpoints\). Transfer this file to your Stable Diffusions Models folder (C:\Users\natemac\AI\stable-diffusion-webui\models).
7. Rename the "last-pruned.ckpt" file to your TRAINING-SAMPLES-NAME "rhaenyra.ckpt".

----------------------------------------------------------------------------------------------
Running Stable Diffusion webui with Trained data:
1.	Run/Double-click on the "webui-user.bat" inside the "stable-diffusion-webui" folder.
2.	Open a web browser and go-to: http://127.0.0.1:7860
3.	In the browser UI goto "Settings".
4.	Under the section "Stable Diffusion" you will see a submenu of "Stable Diffusion checkpoint". Choose your newly created Training Model "rhaenyra.ckpt".
5.	At the top of the page, click "Apply Settings".
6.	Close the web browser & Close the Command Prompt that opened after running "webui-user.bat".
7.	Run/Double-click on the "webui-user.bat" inside the "stable-diffusion-webui" folder.
8.	Open a web browser and go-to: http://127.0.0.1:7860

----------------------------------------------------------------------------------------------
Using Your Trained Data for txt2img:
1.	Using the TRAINING-SAMPLES-NAME & CLASS Prompts to create your image "Portrait of rhaenyra woman".