ErosMix ( SFW & NSFW )版本v1.0 (ID: 135224)

My first experiment with merging models. (combined 34 models). Recipe lower. VAE baked in. My prompts were made with chatgpt by feeding it someone's book on prompting (Recommend this youtube video)

I wanted a good looking model that could do a bit of everything, everything of what I wanted to get (high quality images even with small prompts). The style is: anime semi realistic

- Pros: can do both men and women easy, big breasts, thick/small, fantasy stuff, decent genitalias, did even a dragon pretty decent, tested a few Loras and did alright with them (you gotta test to see if it does fine with ur loras)

- Cons: struggle with sex scenes (unless you're a master at prompting)

My Settings
It should work with your usual settings but here are mine:

• Sampler: DPM++ 2M SDE Karras /Euler a (works with others as well)

• Size: 512-512 /512-768 (I tried 1024-512 and still did pretty fine)

• Clip skip: 2 (did not test clip skip 1 should be alr. ig)

• Sampling Steps: 20-30 (I use 30 all the time)

• Hiresfix upscaler: 4x-UltraSharp /R-ESRGAN 4x Anime6B on 2x, steps: 20, denoise 0.5 (should do fine between 0.3-0.7)

• Cfg scale: 7 (tested up to 15 and was fine)

My Negative prompt: easynegative, badhandv4, lr, beard, curly, veiny, muscular, ((bodybuilder)), facial hair, mustache, lowres, fewer fingers, out of frame, jpeg artifacts, ugly, duplicate, morbid,mutilated, extra fingers, mutated hands, poorly drawn hands, poorly drawn face, mutation, deformed, blurry, dehydrated, bad anatomy, bad proportions, extra limbs, cloned face, disfigured, gross proportions, malformed limbs, missing arms, missing legs, extra arms, extra legs, fused fingers,(speech bubble:1.3),(text:1.3),(logo:1.3),(signature:1.3)

Recipe (left model 70%, right model 30%):

Anyhentai + 1924 = Result1(pruned)

Result1 + abyssorangemix2_hard = Result2

Result2 + akatsukiSoukenV2 = Result3

Result3 + abyssorangemixv2_V2 = Result4

Result4 + amkumix = Result5

Result 5 +animeBabesBiggerABB = Result6

Result6 + anythingsCocoaheaven_v1 = Result7

Result7 + (personal model1 Big_bomber86400) = Result8

Result8 + biggerGirlsModels_v50 = Result9

Result9 + (personal model2 Blondie4736) = Result10

Result10 + CamelliamixNSFW_v11 = Result11

Result11 + cartoonStyle_v02 = Result12

Result12 + cartunafied = Result13

Result13 + chilloutgerph_10 = Result14

Result14 + ChosenChineseStyleNsfw_v10 = Result15

Result15 + chrysanthemum = Result16

Result16 + Destinyglossrev = Result17

Result17 + Divineelegancemix_v6 = Result18

Result18 + duchaitenStylelikeme_v15 = Result19

Result19 + eerieorangemix2Hardcore_eerieorangemix2 = Result20

Result20 + envy2dmix = Result21

Result21 + envyephemeral_v10 = Result22

Result22 + everydayartEda_v5 = Result23

Result23 + forgottenmix_v10 = Result24

Result24 + furnace_09Half = Result25

Result25 + furnaceCu4_v10 = Result26

Result26 + garchazsBasicAnime_garchazmixRealhentai = Result 27

Result27 + grapefruit = Result28

Result28 + grundlemixYoung_v10 = Result29

Result29 + hANYHENFP16_v10 = Result30

Result30 + hassakuHentaiModel_v13 = Result31

Result31 + hentai17 = Result32

Result32 + hentaiflow = Result33

Result33 + HHabemusHentachi_v10 = Result34

(The way I did this is I got 70 models, I combined and tried and then when I merged got new 70 models, I tested them all at once, I did around 30 examples with all 70 at once and the model that got my highest score to my needs was the winner, for example other models did great at sex, but bad at pubic hair or black skin, others had nice genitalia while others had bad genitalias, others had good background while others had bad backgrounds. RIP around $20 to merge this model on runpod q_q)

PS: didn't add any lora concept, I had yet to play with that type of merge

描述:

训练词语: