WebUI: https://github.com/AUTOMATIC1111/stable-diffusion-webui

Variational Auto Encoder (VAE): turn latent to image

Comparison between models: https://stable-diffusion-art.com/models/#Stable_diffusion_v14

Official WebUI Documentation: https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Features


ControlNet: https://huggingface.co/lllyasviel/ControlNet/tree/main/models

AnythingV3(w/VAE): https://huggingface.co/AdamOswald1/Anything-Preservation/tree/main

It seems like only the original 2.x model need config.yml in the same folder

768-v-ema.ckpt, 512-base-ema.ckpt config, 512-depth-ema.ckpt c

.safetensors: a format that is safer (some pickle can be malicious) and faster than pickle (now supported by webui)

// QUESTION: why pickle malicious, what does safetensors do?


Tag Complete: https://github.com/DominikDoom/a1111-sd-webui-tagcomplete

OpenPose Editor: https://github.com/fkunn1326/openpose-editor

Prompt Methods


Sampling Method: With extension, you might need Euler. Otherwise DDIM is good.

Resolution: 768x768 (since it was trained on)

See Miro Board: https://miro.com/app/board/uXjVPChrzFo=/


Miku: 1girl, anime screencap, grey background, pixiv request, pixiv sample, hatsune miku, full body,

Negative: gradient background, (ugly:1.3), (fused fingers), (too many fingers), (bad anatomy:1.5), (watermark:1.5), (words), letters, untracked eyes, asymmetric eyes, floating head, (logo:1.5), (bad hands:1.3), (mangled hands:1.2), (missing hands), (missing arms), backward hands, floating jewelry, unattached jewelry, floating head, doubled head, unattached head, doubled head, head in body, (misshapen body:1.1), (badly fitted headwear:1.2), floating arms, (too many arms:1.5), limbs fused with body, (facial blemish:1.5), badly fitted clothes, imperfect eyes, untracked eyes, crossed eyes, hair growing from clothes, partial faces, hair not attached to head


Table of Content