Use e621 tags (no underscore), Artist tag very effective in YiffyMix.
GridList Species/Artist(v64) update!! & LoRAs (SDXL)/samples/wildcards
Recommend Artist tags (NoobXL) & comfyUI workflow.
Example Setting (Use recommended sampler to get correct quality)
Setting SDXL (SDXL-lightning & NoobXL)
Steps = 12~24
Sampler = "DDPM", "Euler A SGMUniform", "Euler SGMUniform"
CFG scale = 3~4
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
Setting SD 1.5 + vpred
Steps = 30~40
Sampler = "DPM++ 2M Karras", "DPM++ SDE Karras", "DDIM", "UniPC"
CFG scale = 6~8
Negative embeddings = deformity_v6, bwu [SD-WebUI\embeddings]
Stop at CLIP layers = 1
SD VAE = kl-f8-anime2, Furception [SD-WebUI\models\VAE]
Hires. fix
Hires steps = Steps * Denoising strength
Denoising strength = 0.25
Hires upscaler = 4x-UltraMix_Smooth [SD-WebUI\models\ESRGAN]
ControlNet
ControlNet = softedge_hed, control_v11p_sd15_softedge
ControlNet SDXL = softedge_hed, sdxlsoftedge-dexined, noobai-xl-controlnet
ControlNet Weight = 0.35~0.5
ControlNet Pixel Perfect = true
Wan 2.2 ComfyUI Animeted [workflow]
model = wan2.2_ti2v_5B_fp16
vae = wan2.2_vae
lora (turbo) = lora_wan2.2_ti2v_5B_turbo_lora_rank_64_fp16
lora (furry-nsfw) = wan2.2-5B_furry-nsfw-v2.0-e83
lora (livewallpaper) = wan2.2-5B_livewallpaper-720p
sampler = steps:4, cfg:1, sampler:"eular_ancestral", schedule:"sgm_uniform"
SD WebUI
LoRA Training SDXL
imgs count = 15~50
total steps = epoch * imgs count * folder loop = 3000~4500
network_dim = 64
network_alpha = 128 (SDXL) \ 16 (Noob)
learning_rate = 0.0002~0.0005
unet_lr = 0.0001 #learning_rate/2
text_encoder_lr = 0.00005 #learning_rate/4
lr_scheduler = "cosine_with_restarts"
mixed_precision = "bf16"
optimizer_type = "Adafactor"
optimizer_args = [ "scale_parameter=False", "relative_step=False", "warmup_init=False", ]
YiffyMix v4x V-pred Setting
# YiffyMix v4x A1111 SD-WebUI or Forge V-pred Setting
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 )
# YiffyMix v4x ComfyUI or StableSwarmUI V-pred Setting
Add node "ModelSamplingDiscrete", Sampling = "v_prediction"
# ComfyUI V-pred 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]
workflow:[Load Checkpoint (With Config)]-[KSampler]-[VAE Decode]-[Save Image]
# v-pred mode troubleshoot
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.
Version Info
v1.x [2D,512~768] old model, unstable, low resolution
v2.x [2D,512~896] e621 dataset + 30~40% SD5 dataset
v3.0 [2D,3D,512~1088] large dataset then v2.x version, can do 2D, 3D
v3.1 [2D,3D,Real,512~1088]
more realistic, but loss some concept (e621 tag count lower 1000)
v3.2 [2D,3D,Real,512~1088]
unstable version, use SNR version FluffyRock, more noise detail
v3.3 [2D,3D,Real,512~1088]
stable version, more detail, more sensitive to prompts
v3.4 [2D,3D,Real,512~1088] ※include Fluffy Rock Quality Tags-LoRA
stable version, more detail, clear result, reduce some noise (like bush,pattern)
v3.5 [2D,3D,Real,512~1088]
unstable version, contrast & detailed enhance
v3.6 [2D,3D,Real,512~1088]
stable version, more sensitive to e621 tags
v3.7 [2D,3D,Real,512~1088]
stable version, reduce noise overactivity, fix blurry eyes and finger error, fix failed background depth
v4.0 [2D,3D,Real,512~1088]
v-pred 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,Real,512~1088]
v-pred version, color and low contrast fix, reduced realistic noise.
v4.2 [2D,3D,Real,512~1088]
v-pred version, more contrast and detail, yelow issue.
if you feel too fried use Rescale CFG =0.35
v4.3 [2D,3D,Real,512~1088]
v-pred version, clear artist style, fix most yellow/brown issue
the version doesn't need CFG Rescale
v4.4 [2D,3D,Real,512~1088]
v-pred version, no more yellow/brown issue.
the version doesn't need CFG Rescale
v5.0 [2D,3D,Real,896~1536]
SDXL-lightning version, base on Compassmix XL.
v5.1 [2D,3D,Real,896~1536]
SDXL-lightning version, more e621 dataset, lower human face, better NSFW stuff.
v5.2 [2D,3D,Real,896~1536]
SDXL-lightning version, increase average quality. Little unstable than v51 but more creative.
v6.0 [2D,3D,res:896~1536]
More character better sex pose, limited and uncontrollable style (mostlly anime style).
v6.1 [2D,3D,Real,896~1536]
More realistic detail, reduce anime style, fix flat and boring background.
v6.1a-RE [2D,3D,Real,896~1536]
Same as v61 but adjust average style to simi-realistic.
v6.2 [2D,3D,Real,896~1536]
more effective prompt, keep style also with good realistic detail, saturation down little bit.
v6.3 [2D,3D,Real,896~1536]
Detail(noise) level between v62 and v61, upgrade character/artist accuracy.
v6.4 [2D,3D,Real,896~1536]
Upgrade light quality and average detail.
Note about NoobXL to create stable furry (v6x):
use "furry" tag in propmt can let AI create furry not human.
use "no human" tag in propmt can avoid NoobXL model add human and reduce anthro affect (more original style).
use "anime style" in negative propmt can reduce classic booru style and more realistic.
Some SD prompt trick:
Combine two character:
characterA \(characterB\)
Avoid tag bleeding:
(chain:0)-link fence
(cowboy:0) shot
high (collar:0)
Multi-tag combine, enhance and reduce token use:
from side + side view = from side view
crossed legs + legs up = crossed legs up
Basic Style
Negative Prompt SD1.5
unusual anatomy, mutilated, malformed, watermark,
amputee, mosaic censorship, sketch, monochrome
Negative Prompt SDXL
malformed, worst quality, bad quality, signature, text, url
3D Artwork Style
Prompt v5x: blender \(software\), ray tracing, 3d, unreal engine
Photorealistic Style SDXL
Prompt v5x:
by Mandy Disher, by Wim Wenders, by Robert Rauschenberg, [by Fossa666::0.65],
ultra realistic, photorealism, photograph
Negative prompt v5x: sketch, manga, vector, line art, toony
Prompt v6x: film photography, photorealistic, film grain
Negative prompt v6x: anime style, vibrant, pastel
Photorealistic Style SD1.5
Prompt sd15: [:photorealistic, analog style, realistic, photorealism:0.5]
Negative prompt sd15: [:bwu:0.5]
Recommend Refiner: IndigoFurryMix v110-Realistic Switch at 0.5
Description
Recipes v62-Noob:
partA = Chroma XL Mix v331 Mango + Plant Milk Model Suite HempII * 0.55
partB = ZoinksNoob + partA * 0.45
partC = (partB + Shift v10 - ZoinksNoob) * trainDiff:0.45
YiffyMix v62-Noob = partC + LoRA:[SPO-SDXL_4k-p_10ep_LoRA_webui] * 0.55
FAQ
Comments (77)
Do you feel v62 noob is better for anything specific?
Genuinely hope you can bring back some other e621 knowledge, especially for artist styles. Some big artists aren't even showing up disappointingly.
I like the overall look of v62's generations, but the model seems to be very biased towards 3D and realistic styles, therefore the artist styles look almost unidentifiable. The model also lacks recognition for quite a bunch of popular artists
Can you give me an example artist has weaker tags in v62, but works better in early version?
@chilon249 I remember that SD 1.5 versions were good at artist styles, but they are undoubtedly obsolete now. As of V62, I'd say that the styles of whisperfoot, f-r95, photonoko, claweddrip, chelodoy, and many more artists look way better on V60. I really only found one artist whose style works decently enough, and it is dimwitdog
I see youre newest mix uses SDO at 0.55.
This is the cause of people saying the artist tags arent coming through. While this is all experimental and im not 100% sure, I feel its safe to say the following:
You need to carefully merge the OUT blocks. Ive noticed its affects on IN blocks (especially 02 - 05) will begin to poorly affect artist tags.
If you do this very carefully you can push SDO up to 2.2 but its very dependent on how the model will generate images without artist or style tags.
While you really polished Chroma XL better than any one else has. You really need to be careful with ANY models that have chroma XL mixed into it. I would recommend using any other kinda model at the moment in comparison. They made a very messy merge and the weights are so fucked up that it gives lots of tags, signatures, borders and its horrendous with eyes.
Overall though, V62 is much much better than the prior ones and I enjoy gennig with it alot!
Wow. The v62 model does realism GREAT. This is going to be my go-to for realism now because holy shit this is like perfect for me, it does big girls so well
YiffyMix v62 works really well for scale textures I really like it, and it's great for creating overall furry characters most likely going to be my Go-to use for NoobAI model
Hi! I've had an issue for a very long time if anyone could help me please. On my rig I've been able to use 1.5 checkpoints with no issue but for some reason I cannot get any SDXL models to work well with my set up. I've copied meta data and still get very white images when using Comfy UI but when I use Civit AI they are made perfectly? Does anyone know why this is or am I the only one having this problem?
You can try to use forge or reforge UI instead of comfy UI. Install this - https://github.com/LykosAI/StabilityMatrix and then install forge or other UI.
Are they white as in "deep fried"? If so make sure you are using sdxl_vae and not an sd1 vae
@FloatsYourStoat @denis0k They don't come up as fried they're just excessively white with very low resolution. Although I did download Reforge and that seemed to allow XL checkpoints to work. Maybe my SM is having a conflict with comfy. Not sure why or how to fix that. I'll just use reforge for my XL checkpoints for now on. Thank you!
im confused is this a noob model? i thought the old yiffy mix was 1.5.
The older versions are SD1.5. the newest ones are NoobAI
love v62 model!!
Really good AI
It doesn't really like loras all that well. At least in comfy, the only way to correct the green errors is hi res fix.
I lost the v36, and I can't see the history of my arts!
Is v61 vpred while v62;isn't?
How does one add v44 to forge with the config file?
What about merging Illustrious XL 2.0 and SeaArt Furry XL 1.0, i would do it but not practical with merging. However awesome model.
-heterochromia not work?
is it just me or most artist styles look exactly the same (realistic/3d) despite not looking like that in the preview and being clearly cartoony/stylized?
few examples include (BUT NOT LIMITED TO) Bigjazz, Bigcozyorca, B-Epon, Bulumble-Bee, Bustingmangos, Chazcatrix, Commoddity...
I've found that this model in particular has bias for the 3d/realistic-ish type style if you don't specify one, but if you add "flat color 2d toon style" (or something comparable) to your prompt it seems to help bring out the standard toon-ish style that we see in the previews.
But i'm with you - it seems like they're trying to hide the fact it behaves like this, and that annoys me.
This model effectively addresses the issue of blurry backgrounds in noobai illustrations, but it strongly leans toward 3D and realistic rendering without the guidance of an artist tag.
someone already commented this but while these latest noobai model suites are very high quality, unfortunately the newest version as well as 61aRE seem to have a STRONG realistic/3D bias. it seems to REALLY really really really want to generate images in that style even with all sorts of realistic/3D terms in the negatives. hopefully this can be fixed if a new version comes out
What are the recommended resolutions?
SDXL model resolutions:
stable range: 896~1408
simi-stable range: 512~896 or 1408~1920
unstable range: <512 or >1920
by adding extra spo lora in (-0.55) weight can help when you want some style.....
Im usually not a big fan of DPO/SPO or CFG++ kinda stuff.
but he really made the SPO lora work pretty damn well with this model. Its a small sacrifice of creativity for stability.
but you are right. if you want to remove SPO, get the lora and do what you said.
Why is Yiffymix down?
Civitai reduced the number of supported checkpoints from 200 to 100. And apparently, slopmodels for 3DPDs are what people seem to be interested in the most.
@odol wait, seriously? For goodness sake! I was relying on this checkpoint pretty heavily since it's perfect for what I'm currently making...
@odol What's 3DPDs?
@KelvintheNekojin Same
@KelvintheNekojin Same. It's a huge loss, I've been using it non-stop for months now.
@FeralFowler A term of endearment for real-life human females.
Mostly every model I use is stuck behind an auction now. I bidded on a couple, but I friggin hate this. This happened a few months ago too, which drove me to unsubbing. I know somepeople say they want us to try new models, but slamming the door on me using the ones I like does not enourage me to do that.
@BlakjakTheGamer I have no idea why Civitai doesn't realize that continuing their process of enshittification won't exactly encourage people to subscribe to their service. If I ever had any desire to do so, this measure of reducing the total number of available checkpoints at any time certainly dissuaded me.
@BlakjakTheGamer That's not about "they want us to try new models". it's about "They monopolists and want to take all the money they can".
Hope Yiffymix comes back, this was my favorite and most used
I need Yiffymix back! It always happens with WAI Pony too I don't know what's going on
Auction, that's what happen. Waiting till friday.
I used it a week ago, plis Wanna I return work to this model 😩😩😩
dude. plz new sdxl(lighing) model.
n00bAI,pony,illistrious is bad .. cant handle at least 300+ tokens whille SDXL can 400+
This is just plain wrong. SDXL is old and not used anymore for the majority of anime/furry models for a reason. Pictures are messy and the prompt understanding is quite bad in comparison. If you need 300 tokens for a picture, then you should maybe work on your prompts. Guaranteed, there is around 50% of bloat that is worthless.
@anotherstabledif1714 Have you heard anything about small details and control?
Omg, thank you! Thank you so much for bringing Yiffymix back!😍
IT'S BACK!!! HURRAY!!!
Ah. The model didn't win the bet.
I idea maybe creating furry model for wan model base?
He doesn't train models
To anyone who used this model to gen NSFW for free with blue buzz and are now shafted because of Civitai's new policy: you can still gen NSFW with this model for free on Yodayo. Just make an account there (it mentions phone number or Google login but you only really need to give your e-mail address) and fill in your (optionally fake) date of birth and change the content settings in your profile. It only mentions up to R+ rating but the generator doesn't block X or XXX content (you just can't post it). You get 50 free "YoBeans" daily which you can gen pictures with, which cost 4 credits minimum. They don't stack, so you'll have to use them daily so they don't expire. It's not much, but it's better than nothing.
I tried this model Yiffymix v62Noobxl on A1111. But get breast and realistic image even I put them on Negative Prompt. Also, I do not get to use Artist tag. I tried to add artist tag on Positive Prompt but still not work at it.
Oh no! Is it down again?
Congrats on v63!
Something wrong with recipe for v63. What is a partC? Maybe it is a mistake and it is partB instead of partC.
OK, just fix it, is a mistake.
new update :D
When a merge between IL 2.0, Seaart Furry, yiff AI Advanced ?
illustrious v1.0 or v2.0 can't simply merge in Seaart Furry, they are very different SDXL model.
Merge them will create a full-noisy grey goo.
The reason for this v63 update is ChenkinNoobXL (base NoobXL), model has new dataset.
yiff AI Advanced is a stolen model, do not merge
@qek wooow really???
@isperxnovaprime224 Yes, they labeled his weak simple merge as "trained" and started to flood furry sites with his low effort slop and violating ToS on the sites, to advertise his poor paywalled generator
I'm a huge fan of v62. One of the most versatile checkpoints I've tried. The standard, no styling tags, output of v62 is semi-realistic 2.5D for me.
Judging by the preview images of v63, this version is going full 2D, I can't wait to try it out! =)
Even if I don't do furry content, as someone who mainly uses tag based prompts, it can be so useful to have the e621 tag-set as extra tool.
new yiffy lets gooooo
Been waiting for a new version! The example images are phenomenal. One of the few model-makers I know that actually provide variety in samples that show the capabilities of the given model. Never stop cooking. Each new version's very worth it.
We dream and pray of times Chilon blesses us with a new model
I love v62 for being so good and comprehensive, but v52 has a much more encyclopedic knowledge base — it seems to know everything you ask, and its skin detailing is better, too. I hope that someday there will be a model that combines the best aspects of both versions.
Hey, love v63!
The model looks nice and I like it but... the chenkin merge was done wrong to add its knowledge.
You're right, its just still NoobAI 1.1. He didnt do the correct type of merge. Train difference only takes the style really. Add Difference would have been better, but still not get the job done.
it could gen you an image of something later than 2024, but it doesnt have the clip to understand. Therefore theres no way to access that prompt to tell it to make that concept.
Theres a special merge type that allows you to swap or extend the clip, so it has the knowledge of those tag words to create the concept that was merged in.
If you want to see for yourself. Try to prompt a genshin impact character that was released later than March 2025. nefer (30.09.2025) or varesa (19.02.2025) as an example. Or if you know anime better, try using an anime character from an anime that was released after that time line.
Thats how you can tell.
Actually v63 "partA" use addDiff, I forgot to fix recipes info.
v63 has amost 15%~20% ChenkinNoob knowledge, add more ChenkinNoob part will disrupting quality balance.
New character has fewer image, it will be a unstable tag (even in ChenkinNoob), in this case just use LoRA.
Diversity and quality are my first priority, second is accuracy.
@chilon249 I hope you dont think im being offensive or critizing. Theres nothing wrong with the models accuracy. Prompt adherence is great.
As for chenkin, its unstable for some characters, but its more like a gamble if it will get it right.
Add Difference might be suitable as they share the same base model. But having extended clip helps for sure. so XOR or XDARE might be a good option on wieghts below 0.5 for alpha and all 0 for beta.
Atleast this is how I learned about this from Crody (Nova Furry Creator)
Also you are a phenominal prompter, youre cover images show you know how to mix artists really well, along with other stylistic prompts and textures. This model is perfect for someone whom has advanced prompting skills to really see its max potential.
Also it feels like everything worked out for this version. But its something you might want to consider trying for the next version. See if its impact is any greater or worse.
@chilon249 Your v52 model has an impressively large knowledge base, and I’m also amazed by how detailed and natural the skin rendering is in that version. I’ve tried merging v52 with v62/63 in many different ways, but unfortunately couldn’t achieve a satisfying result.
I understand that my request might sound a bit naive, but I still hope it could be possible: having the quality and clean rendering of v62/63 together with the full knowledge depth of v52 would be incredible. Before v62/63 the images tended to look a bit overfried, so the newer versions are a huge improvement — they’re just missing the broad knowledge base of v52.
If it’s technically feasible, it would be amazing if you could consider fully transferring the v52 knowledge into a future version.
Isn't it the oldest model (ID) still getting updates? I literally started with Yiffy (e18)
Can't wait to get to the version that's only 5 of them away!
Details
Files
Available On (3 platforms)
Same model published on other platforms. May have additional downloads or version variants.