Reprint from: https://civitai.com/models/3671?modelVersionId=1876492
Use e621 tags (no underscore), Artist tag very effective in YiffyMix.
YiffyMix doesn't need Pony score tags, Don't use SD1.5 vae in SDXL, you will get fried image.
Example Setting
Setting SDXL (SDXL-lightning & NoobXL)
Steps = 12~24
Sampler = "DDPM", "Euler A SGMUniform", "Euler SGMUniform"
CFG scale = 4
Negative embeddings SDXL = ac-neg1, ac-neg2 (You don't really need this)
Postive LoRA SDXL = SeaArt Quality Tags LoRA (You don't really need this)
Stop at CLIP layers = 2
Hires. fix
Hires steps = Steps * Denoising strength
Denoising strength = 0.25
Hires upscaler = 4x-UltraMix_Smooth [SD-WebUI\models\ESRGAN]
Img2img First Pass ( Upscale enhance )
Scale to = 1.5
Denoising strength = 0.4~0.45
Extra ( seed ) = true
Variation seed = -1 ( use same seed )
Variation strength = 0.9
Img2img Second Pass ( Detail enhance )
Scale to = 1
Denoising strength = 0.3~0.35
Extra ( seed ) = true
Variation seed = -1 ( use same seed )
Variation strength = 0.95
LoRA Training
imgs count = 15~50
total steps = epoch imgs count folder loop = 3000~4500
network_dim = 64
network_alpha = 128 (SDXL) \ 16 (Noob)
learning_rate = 0.0002
unet_lr = 0.0001
text_encoder_lr = 0.00005
lr_scheduler = "cosine_with_restarts"
mixed_precision = "bf16"
optimizer_type = "Adafactor"
optimizer_args = [ "scale_parameter=False", "relative_step=False", "warmup_init=False", ]
Not for commercial purposes.