YuzuLemonTea / README.md
thiros's picture
Update README.md
f43f506
|
raw
history blame
2.28 kB
metadata
license: cc0-1.0
tags:
  - stable-diffusion
  - text-to-image

YuzuLemonTea Mix models ☕

List of my experimental merge models

important notice(Jan 15/23)

According to bbc-mc's note, there is a possibility of bug that some token(prompt) can be ignored, when merge with "add difference" option. Milk and ChaiLatte models are now replaced with bug-fix ver.

https://note.com/bbcmc/n/n12c05bf109cc

Recommended Setteings

VAE: "kl-f8-anime2" and "vae-ft-mse-840000-ema-pruned" are suitable

Steps: 20-30, Sampler: DPM++ SDE Karras or DPM++ 2M Karras, CFG scale: 8, Clip skip: 2, ENSD: 31377, Hires upscale: 2, Hires upscaler: Latent (bicubic antialiased),Denoising strength: 0.54~0.7

Negataive Prompt: (worst quality:2), (low quality:2),inaccurate limb,lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, normal quality, jpeg artifacts,signature, watermark, username, blurry, artist name

  • (worst quality), (low quality) are adjustable between 1.4~2.0
  • If you don't want 3DCG-ish paint, you can add (3d:0.8)~1.0 in Negative Prompt

Sample prompt

4girls,(a 3d reader of:0.8) (teenage loli children:1.2), (wearing intricate casual camisole, cute hair ornament,crop jacket,hot pants, tighhigh:1.1), shiny brown skin, looking at viewer, (alluring smug:1.2), dynamic angle, (onomichi street:1.2),fisheye

YuzuLemonMilk

Block merged model of Anything v3 and some real models. Rather photo realistic. Works fine with positive (realistic) and (photo realistic).

YuzuLemonChaiLatte

Combination of a weight merge of ACertainModel and Anything-V3.0, and a block merge of several realistic models. Rather anime-ish style with realistic background.

YuzuGinger

Add more anime models to YuzuLemonChaiLatte. Can be very anime looks.