>A complete tutorial for Stable Diffusion VAE also called Variational Autoencoder which will help you in fixing desaturated and washed out AI generated images in Stable Diffusion. I will be using Automatic1111 Webui and demonstrate how to bring vae in stable diffusion. The guide gives a complete comparison on different VAEs that are avaliable and how it effects Photorealistic and Semi-Realistic finetuned models. This guide is also helpful in fixing the stable diffusion anime models desaturated colors.
Essential Links (VAE Used):
SD VAE (Original): https://huggingface.co/stabilityai/sd-vae-ft-mse-original
Novel AI/ AnyThing-3.0/ OrangeMix VAE: https://huggingface.co/WarriorMama777/OrangeMixs/tree/main/VAEs
Kl-f8 VAE: https://civitai.com/models/23906?modelVersionId=28569
SD Model Used:
CarDos Anime (Model): https://civitai.com/models/25399/cardos-anime
Dark Sushi Mix (Model): https://civitai.com/models/24779?modelVersionId=93208
PROMPT:
—Copy After This Line—
((best quality)), ((masterpiece)), 1girl, close-up portrait of witch in black dress with fur trim standing in dark forest, frills, purple hair, colored sclera, glowing eyes, witch hat, glowing butterflies, (night:1.2), illustration, realistic, lustrous skin,
Negative prompt: NSFW, (EasyNegativeV2:1), (bad_prompt_version2:1.0), (NG_DeepNegative_V1_75T:1.0), (worst quality, low quality:1.4), (abs, muscular, rib:1.0), (depth of field, bokeh, blurry:1.4), (motion lines, motion blur:1.4), (grayscale, monochrome:1.0), (extra fingers, deformed hands, polydactyl:1.5), large_breasts, text, title, logo, signature
Steps: 30, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 1682867387, Size: 512×768, Model: cardosAnime_v20, Clip skip: 2, Version: v1.6.0
–Copy Before this line—
Please subscribe if you found the video helpful π
Chapters:
00:00 – Introduction & Problems
00:33 – When to use VAE?
01:22 – How to Download VAEs?
01:48 – How to use VAE in Automatic1111 Web UI?
03:52 – VAE Comparison
05:45 – Choosing the Correct VAE
#stablediffusion #aiart #stablediffusiontutorial
