Suzanne's XL Mix
Suzanne’s Mix is made because of a request from Suzanne. I thought it would be a fun challenge. It's not meant for anyone else, but you can download and use it anyway. It attempts to capture some of the composition and aesthetic of RealCartoon-XL by 7whitefire7 and enhance it with some of the more risqué qualities and capabilities of select models. It was much more difficult than I thought it would be, and the earlier versions were too much of a failure for me to post here; therefore, you’ll see that Suzanne's XL Mix starts with version 3
Version 8.0
A new attempt at an old project, I wanted to bring in some refinement and knowledge into Suzanne's model. Version 8.0 is an experiment, but so far, it seems capable of generating some striking images while keeping most of the knowledge of Version 7.0. The newer version generates much more realistic images but still can summon up some of the aesthetics from RealCartoon to add just a bit of surrealism to a scene.
Model Credits
The following models have been or are currently used in a version of Suzanne's XL Mix. Credit goes to the model creators.
RealCartoon-XL https://civarchive.com/models/125907/realcartoon-xl
Hentai ProtoGenesis https://civarchive.com/models/244551?modelVersionId=275930
SDXL Nuclear - General Purpose v2 https://civarchive.com/models/138331?modelVersionId=162051
NSFW POV All In One SDXL LoRA https://civarchive.com/models/144203?modelVersionId=160240
Deep Dark Hentai Mix https://civarchive.com/models/221751?modelVersionId=256648
Pyro's NSFW SDXL https://civarchive.com/models/133808/pyros-nsfw-sdxl
duchaitenReal3dNSFW_v10 - https://civarchive.com/models/247266/duchaiten-real3d-nsfw-xl?modelVersionId=286445
omnigenxlNSFWSFW_v10 https://civarchive.com/models/203014/omnigenxl-nsfw-and-sfw
betterThanWords_v30 https://civarchive.com/models/149359?modelVersionId=233092
FappXL LyCORIS https://civarchive.com/models/248856/fappxl-multi-nsfw-concept
LEOSAM's HelloWorld XL https://civarchive.com/models/43977/leosams-helloworld-xl
Osorubeshi XL (used with permission from natakaro) https://civarchive.com/models/120090?modelVersionId=283551
Description
FAQ
Comments (4)
I have settled on v6.0 for the past 6 months as my primary SDXL checkpoint - better than all the competition IMHO, and there is a lot of it - and now I've done some limited testing of v8.0 and I am tentatively going to say that there is a new Best.
v6.0 has been better than v7.0 mainly because of the eye problems of v7.0 (always way too big no matter how you prompt) but v8.0 appears to have resolved this, bringing v7.0's quality with v6.0's eyes.
You are spot on. I’ve never been happy with v7 as it went too anime, especially the eyes. I’ve tried to make at least 4 or 5 new versions but could never improve on v6…. Until now. There’s still some improvement to be done with v8, as it’s not quite as stable doing HCNSFW as previous versions, but i think the overall potential quality for generation is better than any previous version.
@Dark_infinity there always seems to be a really tough trade-off between HCNSFW and subject quality and stability. After over a year of SDXL, i'm of the opinion that the platform simply can't do it within a single checkpoint. My solution - when i'm putting in effort - has been to do things like generate in AcornIsBoning or NatVis first, then img2img with Suzanne's v6 (now v8), then double back and inpaint any necessary details again with the uglier HCNSFW checkpoint.
@shapeshifter83 It's a good approach. I think it's achievable, but not without significant fine-tuning efforts. I have a raw dataset that could likely achieve some good results, but due to time and processing power restraints, I've never used more than a fraction of it. It'd be pony-level efforts (but with a 100% realistic dataset).