Home » AI Waifu Image General Guide

AI Waifu Image General Guide

AI Waifu Image General Guide

$lave’s note: found on R9k.

/aiwg/ AI Waifu General Local Guide – Automatic1111

Browser alternative


I try to keep it simple as possible – giving you basics and how to get you started.
In case you have further questions, look me under /aiwg/

What you need to know ?

  • Automatic1111 is Stable Diffusion web UI (AUTOMATIC1111 or A1111 for short) is the de facto GUI for advanced users. It enables you to use much more freedom how you use the Stable Diffusion.
  • You can gen on any PC, but I do not recommend anything under 4GB of VRAM (run command “DxDiag” and look under Display settings how much VRAM you have. If you are close to 4000MB you are good, anything above this you are more, than good!)
  • Also less you have VRAM, smaller pictures you can make. Start genning with basic 512×512 px and increase Width and Height. Until you hit out of memory error. In case this happens – close command prompt and start Stable Diffusion again!
  • If you ever get confused about things, I recommend watching videos. Those helped me and will sure help you too.

This is very compressed guide, but if you want to know much more about it, then https://rentry.org/sdg-link is there for you to read. It covers everything about models, Stable Diffusion, LORAs, etc. Also note this guide is covering PC setup and not MAC (for mac look this guide: https://www.youtube.com/watch?v=Jh-clc4jEvk). I also recommend checking: https://rentry.co/waifu-colab-guide out too!

To get started I really recommend looking this video:
https://www.youtube.com/watch?v=kqXpAKVQDNU

To wrap this up:
Note everything is available on: https://github.com/AUTOMATIC1111/stable-diffusion-webui too!

  • Download Python 3.10.6 (later versions do not work!): https://www.python.org/downloads/release/python-3106/
  • Download Git (64bit): https://git-scm.com/download/win
  • Install Python (IMPORTANT!! Make sure you sellect “Add Python 3.10 to PATH”)
  • Install Git (leave everything default and just click Next – Next – Next)
  • Create folder where you want your Stable Diffusion to be.
  • Open that folder and on top where it shows the path of the folder, type: cmd
  • It will open Command prompt
  • Paste: git clone https://github.com/AUTOMATIC1111/stable-diffusion-webui.git
  • Press Enter and wait it to do the thing. (IMPORTANT!!! It can take some time, it can feel it does nothing, but it will on background, so let it finish).
  • Once it is done go into your Stable Diffusion folder
  • Look: webui-user.bat file and open it in Notepad or Notepad++ (what ever you use)

Nvidia – here is example for Low VRAM (4GB and under) users, what you want to have there:

Note --autolaunch will open Stable Diffusion automatically so you do not need to open http://127.0.0.1:7860/ web page manually and git pull will check if Stable Diffusion has new update.

@echo off

set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS= –autolaunch –lowvram –no-half –disable-nan-check –no-half-vae
git pull

call webui.bat

Nvidia – here is example for for Medium VRAM (8GB and under) users, what you want to have there:

Note if you running memory issues just remove “–xformers”

@echo off

set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS= –autolaunch –medvram –xformers –no-half –disable-nan-check –no-half-vae
git pull

call webui.bat

Nvidia – here is example for for High VRAM (12GB and under) users, what you want to have there:

@echo off

set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS= –autolaunch –xformers
git pull

call webui.bat

AMD – users here is what you want to add extra on (+ adjust yourself with –xformers and –lowvram and –medvram if needed):

set COMMANDLINE_ARGS= –precision full –no-half

If you still can’t launch add this line too:

set COMMANDLINE_ARGS= –precision full –no-half –skip-torch-cuda-test

  • And from this point on you are set!
  • To open Stable Diffusion just start: webui-user.bat (I recommend you make shortcut on your desktop for easy access).
  • DO NOT CLOSE COMMAND PROMPT! – it is needed to keep Stable Diffusion running!
  • If it does not open automatically just go: http://127.0.0.1:7860/ web page.

Once you are done you should see something like this (note yours might look bit more different, but we will get you there:

Basic view

  • Go Settings > User Interface and look under “Quick settings list”: sd_model_checkpoint
  • Type two extra things: sd_vaeCLIP_stop_at_last_layers

It will look like this:
Settings view

  • Once it is done, close Stable Diffusion and start it again. And now you see on top: Stable Diffusion Checkpoint (models), SD Vae (your Vaes) and Clip Skip.

Low VRAM users!

Make sure you use Clip Skip 2 or above.

  • Go to Extensions > Available > Load from
    Here are some extensions I recommend:
    Extensions

    • Lycoris – enables you to use Lycoris models
    • Adetailer – enables you to add more detail on your models (recommended: face_yolov8s.pt and hand_yolo8n.pt – add only your negative prompts!)
    • Open Pose Editor – amazing tool for ControlNet
    • Dynamic Prompts – if you want to use Wildcards – video guide: https://www.youtube.com/watch?v=bQK5diN59NA
    • ControlNet – Control how you want your gens be (poses)
    • Infinite Image Browsing – Note there is also Images Browser, but I highly recommend Infinite Image Browsing (as it is faster)!, shows all your images, what you made. !!! I highlighted both on above image – pick one or another !!!
    • Aspect Ratio Helper – helps you to set aspect of the image , by simply clicking one button.
    • Composable Lora and Two Shot – Latent generations (multiple waifus in same image with different prompts). Make sure you enable Composable Lora and Use Lora in uc text model encoder with enabling Latent Couple.

Recommended Models:
(Download models: https://civitai.com/ or https://majinai.art/)

Recommended Vaes:

Recommended Textual Inversions embeddings:

Recomended Upscalers:

Luna and Saeriah

 

Composable Lora and Two Shot
Model: AnyLora
Vae: Pastel Waifu Diffusion Vae
Sampling Method: DPM++2M Karras
Adetailer: face_yolov8s.pt and hand_yolo8n.pt
Upscaler: 4xUltrasharp

Step by step how I myself gen:

  1. Pick desired Model and Vae
  2. Clip Skip: 2
  3. Write prompts
  4. Enable ADetailer using both face_yolov8s.pt and hand_yolo8n.pt – copy paste under both of them only your negative prompts
  5. Sampling Method: DPM++2M Karras
  6. CFG Scale: 7
  7. Sampling Steps: 20 (if it is Hyper Realistic Model I use 30-40)
  8. High Res Fix is nice, but not needed (you do not really needed it, as ADetailer and Upscaler does much better job)
  9. Generate
  10. Once image is generated click “Send to Extras”
  11. Resize:1 (what ever you like)
  12. Upscaler: 4x-Ultrasharp
    1. In case you want more detail on images, then after Generating send image to img2img
    2. Keep the same resolution
    3. CFG Scale: 7
    4. Denoise Strength: 0.25 – 0.35 (you need to find what you like, but note as you change this, more your image will change – I found 0.25 is sweet spot)
    5. Go under scripts and look SD Upscaler
    6. Tile Overlap: 64
    7. Scale Factor: 2
    8. Upscaler: 4x-Ultrasharp
    9. Generate

// Luna anon

Original Article 

/aiwg/ AI Waifu General Colab Guide (I’m terrible at writing guides edition)

 

  • Collab is kind of unstable while running it. Sometimes things will just crash and you’ll have to restart everything. Make sure to save the file separately, with your changes. Most of times, restarting the Start Stable Diffusion Web UI cell works though.
  • You can also just refresh the page if it seems the generation stopped happening and no error happened.

This is a simplified guide made by someone who can’t explain things right. However, if you’re interested on the topic of what are those models, Stable Diffusion, LORAs, etc, you can go on Stable Diffusion general on /g/ or go https://rentry.org/sdg-link for the long stuff.

To get started, open the following link in your browser: https://colab.research.google.com/github/Linaqruf/sd-notebook-collection/blob/main/cagliostro-colab-ui.ipynb. Once you have the notebook open, you’ll need to add some custom URLs for the models you’ll be using on Custom Download Corner cell. Copypaste the suggested ones below or search them anywhere (Civitai, HuggingFace, etc).

custom_model_url section:

https://civitai.com/api/download/models/30163, https://civitai.com/api/download/models/25644, https://civitai.com/api/download/models/25472, https://civitai.com/api/download/models/44686

custom_vae_url section:

https://huggingface.co/hakurei/waifu-diffusion-v1-4/blob/main/vae/kl-f8-anime2.ckpt, https://huggingface.co/samle/sd-webui-models/resolve/main/pastel-waifu-diffusion.vae.pt

custom_LoRA_url section:

https://civitai.com/api/download/models/18043, https://civitai.com/api/download/models/15603, https://civitai.com/api/download/models/18765

custom_embedding_url

https://huggingface.co/datasets/gsdf/EasyNegative/resolve/main/EasyNegative.safetensors, https://huggingface.co/AsciiP/badhandv4/blob/main/badhandv4.pt, https://huggingface.co/datasets/Nerfgun3/bad_prompt/blob/main/bad_prompt_version2.pt

Once you’ve added these URLs, you can run the cells at order as below

  • Install webui
    Imgur
  • Download standard models and vae
    Imgur
  • Install Controlnet
    Imgur
  • Install custom stuff (as given above)
    Imgur
  • Run cagliostro web ui
    Imgur
  • Once it runs, some urls will appear. You can click on the one whose the colab recommends the most in order (ngrok > gradio > cloudflared > remotemoe > localhostrun > googleusercontent).
  • If you want better stability during your gens, I recommend running as ngrok. You’ll need to create an account though.

If you want more information about prompts and other details, check them out.

If you feel lost on what to gen, or how to gen, start easy. Try to gen the girl design first and later try different poses and situations.
Below are a piece of the models I’m playing around and prompts.

Hornette Donusteel (on cockpit)

Image description

  • Positive prompt: 
    <lora:SecretSauce610-fp16-pruned:0.3>, 
<lora:Light and Shadow:0.5>, 
<lora:blacklight:0.5>,

[[INSERT YOUR WAIFU PROMPTS HERE]]

(cockpit, piloting, spacecraft, sitting, thighs, straight-on, airplane interior, cable, wire, seat:1.5),
(purple theme, orange theme, rain, storm, thunder:1.4), (patterns),
  • Negative prompt: 
    (worst quality, low quality:1.4)
  • Settings: 
    Steps: 22, 
Sampler: DPM++ 2M Karras, 
CFG scale: 7, 
Seed: 1414689898, 
Size: 768x768, 
Model: velaMix_Version1, 
Denoising strength: 0.55, 
Clip skip: 2, 
Hires upscale: 1.5, 
Hires steps: 10, 
Hires upscaler: SwinIR 4x

Gorg (high contrasted)

Image description

  • Positive prompt: 
    [[INSERT YOUR WAIFU PROMPTS HERE]]

(bright pupils, outline, high contrast, neon trim, neon palette, black background:1.5), (musical note, music, shaded face:1.3), 
(((pixel art))), sprite, retro, (retro videogame),
  • Negative prompt: 
    (worst quality, low quality, multiple girls, 2girls:1.4), easynegative,
  • Settings:
    ⎗
    Steps: 22, 
Sampler: DPM++ 2M Karras, 
CFG scale: 7, 
Seed: 1993881507, 
Size: 768x768, 
Model: bluemixTMNDEnhanced_v10, 
Clip skip: 2

    Source

Superintelligence: Paths, Dangers, Strategies

Superintelligence: Paths, Dangers, Strategies

The Singularity is Near

The Singularity is Near

Do Your Part and Report Illegals To ICE

Do Your Part and Report Illegals To ICE

Project 2025’s Mandate for Leadership: The Conservative Promise

Project 2025’s Mandate for Leadership: The Conservative Promise

Morella

Morella

The Unnamable

The Unnamable

Leave a Reply

Your email address will not be published. Required fields are marked *