Use e621 tags (no underscore), Artist tag very effective in YiffyMix v2/v3 (SD/e621 artist)
Species/Artists grid list (update v36) & Furry LoRAs/samples/wildcards
VAE:WD-KL-F8-Anime2, Furception correct contrast and noise reduction [Settings\Stable Diffusion\SD VAE]
ClipSkip:Stop at CLIP layers=1 (SD=1, NAI=2) [Settings\Stable Diffusion\Clip skip]
v4.x is vpred version, need download ".yaml config" with model, see version info.
recommend YiffyMix v4.x use refiner [0.3~0.6] with other EPS (non-vpred) Model.
Non vpred version recommend v34 & v36
# Some problems #
If you use new version webui-forge and failed detect v-pred model.
Try update webui-forge (click update.bat).
When loading v-pred model, you will see this line in cmd console:
left over keys: dict_keys(['v_pred'])
# v4.x some time will get little fried issue image in this case #
Use "Dynamic Prompts with __wildcards__ prompts" and "batch size > 1"
Just generate again with same prompts with parament, it will return normal.
# V-pred SD WebUI Setting #
YiffyMix 4.x is v-prediction model, need v-pred ".yaml" config:
1.Download config file ".yaml" paste next to the model.
2.Raname ".yaml" same as model name. ( check the ".yaml" has [parameterization: "v"] line )
3.Restart SD-WebUI ( if config fails to load, the model just generate noise )
# V-pred Comfy UI Setting #
1."ModelSamplingDiscrete", Sampling = "v_prediction" [Right Click\AddNode\advanced\model]
[Load CheckPoint]-[ModelSamplingDiscrete]-[RescaleCFG]-[KSampler]-[VAE Decode]-[Save Image]
# V-pred Comfy UI Setting (old version) #
1.Copy config ".yaml" to [ComfyUI\models\configs] and refresh ComfyUI
2."Load Checkpoint (With Config)" [Right Click\AddNode\advanced\model_merging]
[Load Checkpoint (With Config)]-[RescaleCFG]-[KSampler]-[VAE Decode]-[Save Image]
Example Setting
Txt2img Setting
Steps = 40~50
Sampler = DPM++ 2M Karras, DPM++ SDE Karras, DDIM, UniPC
CFG scale = 5~10
Negative embeddings = deformity_v6, bwu ※[SD-WebUI\embeddings]
Hires. fix
Hires steps = 15
Denoising strength = 0.45~0.5
Hires upscaler = 4x-UltraMix_Smooth ※[SD-WebUI\models\ESRGAN]
Img2img Resize
Scale to = 1.5~2.0
Denoising strength = 0.45~0.5
ControlNet0 = softedge_hed, control_v11p_sd15_softedge
ControlNet0 Weight = 0.4~0.5
ControlNet0 Pixel Perfect = true
ControlNet1 = Reference
ControlNet1 Weight = 0.2~0.3
ControlNet1 Pixel Perfect = true
SD Setting
Stop at CLIP layers = 1
Eta noise seed delta = 0
SD VAE = kl-f8-anime2 ※[SD-WebUI\models\VAE]
Txt2img Animated Diff ※no "BREAK", tokens < 75
Steps = 30~50
Sampler = DDIM
CFG = 10~20
ImageSize = 512~768 ※large image less motion
Number of frames = 10~14
Negative Prompt v2
unusual anatomy, pixel \(artwork\), mutilated, malformed, watermark, amputee,
restricted palette, mosaic censorship, deformity, sketch, unfinished, monochrome
Version Info
v1.0 [2D,res:512~768]
old model, unstable, low resolution
v2.x [2D,res:512~896]
e621 dataset + 30~40% SD5 dataset
v3.0 [2D,3D,res:512~1088]
large dataset then v2.x version, can do 2D, 3D
v3.1 [2D,3D,Realistic,res:512~1088]
more realistic, but loss some concept (e621 tag count lower 1000)
v3.2 [2D,3D,Realistic,res:512~1088]
unstable version, use SNR version FluffyRock, more noise detail
v3.3 [2D,3D,Realistic,res:512~1088]
stable version, more detail, more sensitive to prompts
v3.4 [2D,3D,Realistic,res:512~1088] ※include Fluffy Rock Quality Tags-LoRA
stable version, more detail, clear result, reduce some noise (like bush,pattern)
v3.5 [2D,3D,Realistic,res:512~1088]
unstable version, contrast & detailed enhance
v3.6 [2D,3D,Realistic,res:512~1088]
stable version, more sensitive to e621 tags
v4.0 [2D,3D,Realistic,res:512~1088,VAE Baked]
v-pred stable version, remix from EasyFluff
accuracy anatomy and style with fewer prompt,
little dim blue average color with smooth noise, weakly negative prompt issue
v4.1 [2D,3D,Realistic,res:512~1088,VAE Baked]
v-pred stable version, color and low contrast fix, reduced realistic noise.
v4.2 [2D,3D,Realistic,res:512~1088,VAE Baked]
v-pred stable version, more contrast and detail.
if you feel too fried use Rescale CFG =0.35
(SD-Forge has builtin, LatentModifier Integrated/Rescale Cfg Phi)
# V-prediction model has following feature #
Low-Rescale CFG : low frequency noise, cluster detail, high contrast
Sometimes less detail but smooth surface color and less mutation.
High-Rescale CFG : high frequency noise, flat detail, low contrast
more detail can be shown, but not like EPS (non-vpred).
You can see the visible halftone dot in "starry night sky" or "grassland".
try "halftone background" to test it.
ArtStyle Demo (use Dynamic Prompts)
{1-4$$by Ben Hatke|by Ed Mell|by Tomer Hanuka|by Syd Mead|by Nicholas Roerich|by Anton Fadeev|by Franz Kline|by Bigcozyorca|by Inkplasm|by Retros}
<Line Art v1>
{1-4$$by Brian Kesinger|by Darwyn Cooke|by Mike Mignola|by Winslow Homer|by Gabriel Ba|by Cumbread|by Edtropolis|by Buta99|by Claweddrip|by Domovoi Iazaroth|by Donkeyramen|by Haps}
<Line Art v2>
{1-4$$by Alphonse Mucha|by Katsushika Hokusai|by Dan Mumford|by Julie Bell|by Jerro|by Lawyerdog|by Rapscallion|by Twinkle-Sez}
<Line Art v3>
{1-4$$by Faustsketcher|by Jessie Willcox Smith|by Albert Marquet|by Cyril Rolando|by TheSecretCave|by Patto}
<Special>
{1-4$$by Android Jones|by Harry Clarke|by Leonid Afremov|by Artur Bordalo|by Marsminer|by Sagaris Uwu|by Rabbity|by Ralek}
Painting Art:
{1-4$$by Wlop|by Vincent Van Gogh|by Robert McCall|by Hermenegildo Anglada Camarasa|by Joaquín Sorolla|by Supplesee|by Kenket|by Kenno Arkkan}
<Anime>
{1-4$$by Rumiko Takahashi|by Ephraim Moses Lilien|by Yuumei|by Ken Sugimori|by Makoto Shinkai|by Hioshiru|by Fumiko|by Ricegnat|by Zangusuu|by Glacierclear|by Reysi|by Slugbox}
<Fluffy Fur>
{1-4$$by Ferdinand Knab|by Reylia Slaby|by Larry Poons|by Fernand Khnopff|by Harriet Backer|by Anchee|by Foxovh|by Lostgoose|by Sabretoothed Ermine|by Silverfox5213|by The Gentle Giant}
3D Artwork Style
Tip:
(3d \(artwork\):1.1~1.5), (unreal engine:1.1~1.5) = "3D style" weight tweaker
Prompt:
###put some prompt here###
BREAK
(masterpiece, best quality, ultra realistic, 4k, 2k, (intricate:1.1), (high detail:1.2),
(3d \(artwork\):1.15), blender \(software\), {1$$sharp focus|soft focus}, ray tracing, (unreal engine:1.3), absurd res)
Photorealistic Style
Tip:
(intricate:0.8~1.1), (high detail:1.1~1.5) = "Realistic detail" weight tweaker
Prompt:
[[:(by Chunie:0.85), by Meesh, by Silverfox5213, by LevelViolet, by Todex, by Jackson Pollock:0.16]::0.66],
###put some prompt here###
BREAK
(masterpiece, best quality, ultra realistic, 4k, 2k, (intricate:0.9), (high detail:1.4), film photography, {1$$sharp focus|soft focus},
RAW photo, photorealistic, analog style, subsurface scattering, photorealism, absurd res)
Negative prompt:
deformity_v6, bwu, high contrast, over saturated, glossy, cartoon, toony