V4
Completely redesigned the logic and architecture of the model. It can be said that it is almost a new model not similar to previous merges, but it contains almost everything the same as the previous versions, with some exceptions.
Sampling method - Euler a
Schedule type - Normal (Automatic)
Sampling steps - 30
CFG Scale - 6
Prompt:
TAGS + NL
very awa, newest, recent, masterpiece, best quality, absurdres, highres, detailed, Neg. prompt:
worst aesthetic, worst quality, low quality, old, early, mid, ai-generated, ai-assisted, bad anatomy, artistic error, lowres, bad hands, signature, artist name, scan artifacts, jpeg artifacts, unfinished, watermark, logo, text, dated, username, web address, bad feet, censored, blurry, off-topic, bad proportions, bad perspective, what, meme, tagme, copyright name, resized, downscaled, source larger, ugly, aliasing, cropped, MD5 mismatch, moire, third-party edit, image sample, upscaled, third-party watermark, has bad revision, 4-koma,V3
Sampling method - Euler a
Schedule type - Normal (Automatic)
Sampling steps - 30
CFG Scale - 7
Prompt:
very awa, masterpiece, best quality, year 2024, newest, highres, absurdres, very aesthetic, amazing quality, high resolution, ultra-detailed, detailed, score_9, Neg. prompt:
S:
worst aesthetic, good quality, normal quality, low quality, bad quality, worst quality, mid, early, old, adversarial noise, ai-generated, ai-assisted, stable diffusion, lowres, low res, monochrome, clone, XL:
very displeasing, displeasing, bad anatomy, artistic error, lowres, bad hands, signature, artist name, multiple views, artist sign, worst aesthetic, worst quality, low quality, bad quality, scan artifacts, jpeg artifacts, lossy-lossless, ai-generated, stable diffusion, abstract, official art, old, 4koma, reference sheet, dakimakura \(medium\), turnaround, comic, greyscale, monochrome, sketch, unfinished, watermark, logo, text, dated, username, web address, bad feet, extra digits, fewer digits, extra arms, extra faces, multiple heads, missing limb, amputee, severed limb, mutated hands, distorted anatomy, worst detail, censor, worst aesthetic, good quality, normal quality, mid, early, adversarial noise, ai-assisted, low res, thumbnail, compression artifacts, censored, bar censor, mosaic censoring, blurry, off-topic, bad proportions, bad perspective, what, meme, tagme, copyright name, copyright symbol, resized, downscaled, source larger, ugly, aliasing, simple background, asymmetrical, concept art, cropped, bleed through, crease, bad aspect ratio, derivative work, MD5 mismatch, moire, third-party edit, image sample, upscaled, Waifu2x, unmoving pattern, third-party watermark, has bad revision, doesnotexist, bkub, Recipe:
Final = [
[
# Left branch
{
chromaxlMix_v331 + indigoFurryMixXL_noobaiEPS [α=0.5]
} + {
novaFurryXL_illustriousV40 + rouwei_v07Epsilon [α=0.5]
} [α=0.5]
] + [
# Right branch
{
[
ntrMIXIllustriousXL_xiii + voivodspell_v10 [α=0.5]
] + [
4thTailAnimeHentai_050 + hassakuXLIllustrious_v2 [α=0.5]
] [α=0.5]
} + {
waiNSFWIllustrious_v110 + [
ponymagine91_gamma + (
ponymagine91_gamma - animagineXLV31_v31 + waiNSFWIllustrious_v110 [α=1.0]
) [α=0.5]
]
} [α=0.5]
] [α=0.5]
] + [
Illustrious-XL-v1.0 + (
noobaiXLNAIXL_epsilonPred11Version - cyberillustrious_v16 + illustriousXL_v01 [α=1.0]
) [α=0.5]
]V2
What's new in the version?
Added normal Pony tag support.
Improved anthropomorphic anatomy.
Added ability to make good photorealistic generation.
Added:
waiNSFWIllustrious_v10
AnimagineXL v40
chromaxlMix_v331
bigLove_xl1
Prompt:
very awa, very aesthetic, masterpiece, best quality, amazing quality, newest, recent, absurdres, high score, great score, score_9, score_8_up, score_7_up, Neg. prompt:
worst aesthetic, good quality, normal quality, low quality, bad quality, worst quality, low score, bad score, average score, mid, early, old, Krabby Patty Model secret formula:
=== Base Model ===
base_model = noobaiXLNAIXL_epsilonPred11Version=== Reinforcing the "Illustrious" style. ===
illustrious_enhancement = waiNSFWIllustrious_v100 + ((cyberillustrious_v16 + bigLove_xl1) - illustriousXL_v01)=== Fine Tuning Styles (Anime/Pony & other mixes) ===
style_tuning = (((animagineXL40_v40 + ponymagine91_gamma - animagineXLV31_v31) + autismmixSDXL_autismmixPony - ponyDiffusionV6XL_v6StartWithThisOne) + (chromaxlMix_v331 - noobaiXLNAIXL_epsilonPred11Version))=== FINAL RECIPE ===
final_recipe = base_model + illustrious_enhancement + style_tuningnoobaiXLNAIXL_epsilonPred11Version
+ { [ waiNSFWIllustrious_v100 + (cyberillustrious_v16 + bigLove_xl1 – illustriousXL_v01) ]
+ [ ( (animagineXL40_v40 + ponymagine91_gamma – animagineXLV31_v31)
+ autismmixSDXL_autismmixPony
– ponyDiffusionV6XL_v6StartWithThisOne )
+ (chromaxlMix_v331 – noobaiXLNAIXL_epsilonPred11Version) ]
}
V1
BE SURE TO USE VAE SDXL
I WAS ABLE TO MERGE
PONY XL
ANIMAGINE XL
WAI ILLUSTRIOUS
CYBER ILLUSTRIOUS
NOOB EP
IT'S SUCH A HELL OF A MIX, BUT IT WORKS... AND IT WORKS WELL...
Model supports LoRA for all of the above models, making it extremely versatile.
These prompts are provided as an example and are based on the merge models. Feel free to modify them as you see fit.
Prompt:
very awa, very aesthetic, masterpiece, best quality, amazing quality, newest, recent, absurdres, score_9, score_8_up, score_7_up, Neg. prompt:
worst aesthetic, worst quality, worst detail, low quality, bad quality, normal quality, old, early, mid, Recipe: NOOB + ((WAI IL + CYB IL - IL) + (PONY + ANIMGN) - SDXL)
Description
FAQ
Comments (8)
Compared to V1, I found that the V2 has:
- weaker, less detailed backgrounds
- stronger default style at the cost of weaker adherence to style prompts (must bump CFG/weights to make it draw 2D images unaffected by pseudo-3D)
- less vibrant/contrast compared to V1
If I may suggest, would you consider making another merge without any photorealistic checkpoints in the mix (e.g. bigLove)? I suggest using the following checkpoints in the merge:
For Animagine:
Animagine 4.0 or maybe
knkKirimagineAnimagine_v10 as more stable (or both): KNK Kirimagine (Animagine v4) - v1.0 | Stable Diffusion XL Checkpoint | Civitai
For Pony:
4th Tail 0.5 for diversity (big Pony finetune) 4th tail (anime/hentai) - 0.5.0
For Illustrious:
RouWei 0.7 EPS for diversity RouWei - v0.7 epsilon | Illustrious
WAI to stabilize output WAI-NSFW-illustrious-SDXL - v10.0 | Illustrious
WOW, thank you so much! That will help a lot!
Hi, please try the new V3 version and leave a comment!
@voider Downloaded already, first results look encouraging; it clearly beats the V2 already. I will try some more and post a few pictures and a short review tonight.
@civit77899 Thank you very much! I'll be waiting!
[TLDR: V1.0: Better than 2.0 Imo. V1.0 has crazy good LoRA absorption and a nice fluid base style.]
This merge is simply insane. It absorbs even the most over/underbaked LoRAs effortlessly, adheres to and blends artist tags perfectly, and stages aesthetically pleasing compositions better than most Illustrious merges here. The only other model that comes to mind for this level of ease of prompt-to-composition adherence is "Mistoon Cube' (Though Mistoon Cube is fairly locked in its style).
As for V2.0, I find its LoRA absorption worse than V1.0. That's not to say it's "bad", as it's on par with most other models, but that V1.0 is just that much better when compared (may be personal preferences). Sure, you can get good LoRA results with base Illustrius models (vanilla "illustrious" and "RouWei), but you usually sacrifice a base style for more LoRA fluidity. With Chimera, you get the best of both a base and a styled model.
I can't wait to see where this goes with V3.0 and beyond.
Hi, thanks for the comment!
Alas I'm stuck on the merge pony with anything. It just doesn't want to work.
The answer to why v2 is worse is very simple - the admixture of a realistic big love model, I did it for nothing.
Hi, please try the new V3 version and leave a comment!
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.



















