EDIT II: Here is my latest NSFW realistic version, it came out of a mix of my illustrious-real and a noobai model I'm working on, it gave impressive results at our testing, so here I share it to you !!!
EDIT: I just added my new Comfyui Workflows as 2nd file, it shows training data but its the workflow I use.
THE HYPER-SD VERSION IS NOT REALLY AN HYPER-SD, TRY IT WITH YOUR USUAL SETTINGS IT GIVES GREAT SURPRISES!!!!
got my Ko-Fi account: https://ko-fi.com/ziggart If you like my work a coffee is always welcome ! :)
###EDIT 2###
Dear User, I heard you,
some like more the pony versions, like 22.6 or 23.5
Many of you prefer the realistic verions like 19.5 or 20.57
That is why I will separate both branch, this one will become the pony one, I will soon post an hyper-sd version of 20.57, in a new post, its ready I just need to do my samples.
So the 20.57 branch is now here : https://civarchive.com/models/426082/mklan-xxx-nsfw-realistic
###End of edit 2###
####EDIT --- IMPORTANT NOTE MODEL Mklan23.5XXX2StepsHSDDPOTurbo ####
This model is merge with HYPER-SD LORA, IT MAY NOT WORK YET ON FORGE OR A1111
it give nice draft result at 2 steps - good result at 20 to 35 steps - and incredible result around 50 steps - Always with 1 - 2 CFG - Samplers I use is DPMPP-SDE with Scheduler SGM_Uniform
#####END EDIT#####
General Notes
Configuration:
Optimal Settings: See the 'About this version' tab on the right for specific information on each version.
Typical Configuration Details: CFG: 2 | steps: 10 | DPM++ SDE
(Samplers and schedulers can be adjusted based on personal preference.)
Engine Used: Samples were created with ComfyUi, but have been tested with other systems like A1111 and Forge.
See my ComfyUi Workflow here: https://civarchive.com/articles/4338
Connect:
A special thank you to my followers who joined our Discord for testing and feedback. Your support is invaluable. Come join the server to discuss models and share generations.
Discord Link: https://discord.gg/MABJUA8tcF
Streaming Update: Following a ban from Twitch, I'm moving to Kick. Join here: https://kick.com/ziggart
Share and Review: Please share your generations to the model page and write a review as it helps me understand how to better rollout future versions.
Version specific
V22.5XXX - The version of Reason - A1111, Forge, Fooocus and Comfyui compatible !
we believe issue has been fixed, this model can do crazy things now.
Remember it is first destined to do sexual act, but now it can do much much more!
V21.6XXX Version - Happy Easter!
This version has shown great results, and I'm excited to share it with you.
V21
It seems v21 is more sensible to cfg / steps and sampler/scheduler changes.
Use 'FreeU' and 'Self Attention Guidance' for even better results.
VAE Included: All image samples are provided as is—no refinements, upscaling, retouching, not even negative prompting. (which is why my samples have watermarks)
V20
V20 is out, thanks to my followers who joined Discord and helped testing all the tryouts!
This version introduces a realistic blend of various styles, which includes ponys fapp, pyro, nsfwallinone, achieving a difficult balance that sets it apart from the exceptional 19.51 version.
Pony style loras: Try some of the prompts from the loras included below in my models. If you turn on lora it will add some style but it will keep the image more realistic instead of animation.
To prompt efficiently with this model, you can copy Pony prompts without the 'score_#' parts.
V19
After long deliberation, I decided to release my XXX merge.
for my samples I used cfg 2 - 15 steps - dpmpp-2m-sde - Exponential (corrected, the 2s-ancestral + karras gave a white dot thingy...or maybe it was a bad vae in the merge, corrected since turbo version)
I also tested this prompt from Seeker70 , as I merge his Acorn is Boning in my last merge:
highest quality photograph of beautiful naked 28 years old woman {(vaginal sex:1.5) with man, erection |(from above, BLOWJOB, big cock) | (manually masturbating:1.3) with (blue glass dildo insertion:1.3) | (kneeling), (receives cum in open mouth from man's cock:1.4) | (squatting, peeing piss:1.4) | (licking penis, cum:1.4) | (giving deep throat blowjob:1.4) | (having sex with another woman:1.), (manual dildo inserted:1.5) | (POV BLOWJOB:1.4) | (having anal sex:1.4) | (kneeling masturbation, touching own pussy:1.4)} {in the bathroom | in hot tub | in the kitchen | by the pool | in bedroom | on rooftop in the city}, {ginger | blonde | brunette} {long | medium | short} hair, {petite | skinny | chubby} body, {(large breasts:1.3) | (medium breasts:1.3) | (small breasts:1.3) | (huge breasts:1.3)}, (intricately detailed skin), (perfect lips), high contrast
Special thanks to Artvandelay on Discord who rewrote this for me.
Description
After several merge, trials and error, help with my follower on discord to test all iteration and several mistake, Mklan 20.5XXX is finally here so yes it's v20.57XXX that show that it did not work properly on the first 20.5xxx.
Also, I fixed on that version all clip skip or VAE error, all is included, no need refiner or else, and it should work fine on all system (sd1111, forge or comfyui)
For samples I used DPMPP-2S-Ancestral - Exponential - CFG 2 - 10 steps
FAQ
Comments (48)
Odd. When I try to download 20.57 I just get a file called pricing.htm.
It is lock until tomorrow , in the meantime it's available for subscribers or on discord
@mskiller51 Thank you. I thought there might be something wrong after the downtime today.
The best XXX model yet. At least not everything is "close up" here, you can do full scenes with it.
Thank you for the review !
One of my favorite. Can anybody answer, why models sometimes looks like cartoon characters? :)
well because in the merge there is some anime involved in order to get as many position, tricks as possible..
Lmao no way you just updated... 2 mins ago I saved my preset. After 2hrs of testing with v20.57
well it took me way longer to baked it, as we did test several version on my discord for the last 2 days ... ;) ...usually when my realistic version comes out, then comes the art, then the hentai, then the xxx...then anime, then allinone, and finally i merge them with lightning for their lightning version, if you like to play with my model, join us on discord! ;) https://discord.gg/Eq2DXWsJ
Has anyone gotten this to work in Fooocus? The recommended settings do not seem to work.
I just downloaded and checked in Fooocus two versions:
- 1951xxxturbo works even with default settings
- 215 XXX produces a bunch of artifacts in the image (as if Photoshop increased the noise, and then turned on the Gaussian blur filter).
settings were used:
1. by default
2. described by the author (+ version edited by the author)
3. Custom (with various variations)
Will do some test, maybe a fix will come today
@Maximo666 @gltchcom @v51761343901 I just posted some sample on how to make it works on Fooocus, I never use that one, too much restriction, I suggest you check Forge, its as easy but give you all capability of comfyui, you got more speed and better quality...
@mskiller51 I may have misunderstood (English is not my native language), but I read that you managed to run your model in Fooocos, and you showed how to do it somewhere. I just don't understand - where? (I'm aware of other SD interfaces, thanks)
Everything was found! )))
Thanks!
@v51761343901 you're welcome, as I do not cheat on my samples, I try to help out every one getting same result as i do. for info, my comfyui workflow include FreeU and Self Attention Guidance, which are I believe not available on fooocus or A1111, which is why I strongly suggest Forge for people who are scared of nodes or comfyui for the real ones ;)
I can't find the settings
@lorenzopaolicchi89678 its no more there, as the model that has issue is removed, it should works fine although fooocus is a noob crap, better use forge or comfyui ..
@mskiller51 if you do not cheat on your samples then why is it so unclear and inconsistent on how to replicate them? why do so many people have so many problems? at a certain point something is dishonest OR you are using precise settings that you are not CLEARLY and EXACTLY explaining to people. give every parameter, sampler, steps, cfg scale, clip, resolution, engine, etc.
so far, there is a big inconsistency between your claims and what people are able to do. it cant be magic or entirely user error, so i dont want to call you a liar, but maybe a bad communicator.
@captainearlgreyhot449 Like said stop your bullshit on all conversation come over discord we will show it to you live, then you stop crying because you don't jnow how to use it!
Unable to get this one to work. It just creates deep-fried images. It's clearly over trained/fitted to something. Using authors settings.
Which system do you use, I use comfyui, but it seems to have issues with fooocus, I am looking to fix this.
@psspsspsspssspss Do you see the results in the gallery? Should give you a hint where to fix your problem.
@psspsspsspssspss I just posted some Fooocus result , no need to put a down on your comment, your issue is Fooocus, its a too much noob system that assist so much that it does not let you enters the proper value, I added in my gallery how to set it up ! there you go...
@jtk1996606 Thank you for your support ;)
@mskiller51 I'm not using fooocus :/ I'm using the normal auto1111 webui.
@psspsspsspssspss try forge, or wait fix is coming...I hope
@psspsspsspssspss new version should work now..
I have the same problem of getting fried and noisy images using the recommend settings in v21. 5.
I'm using comfyui!
The previous Version workes just fine.
What helped so far was reducing the steps to a value of 5. The result is still slightly grainy though.
you can download all my img as png, my workflow is embedded, no cheating here, but i use self guidance to help out
@St1g check the new version ...
Trying this out in Automatic 1111 and getting very grainy, messy images (but so close to good). What am I missing?
try Forge, add FreeU and Self Attention Guidance. I did not try it on auto1111, will do some test maybe later.
@mskiller51 ok... let me know if you do because auto is all I use and would love to get this to work
Actually preparing a version 21.55xxx to solve all issue, hope it will be the case..
new version is out, did you try ? @SDXLWizard
@mskiller51 will try later this evening and let you know...
@mskiller51 Using 21.59 and still everything is pretty grainy and fried. Euler a is the closest to a good generation but still spotty. Auto1111, btw
@danthantanos still did not try with a111 myself, did some live yesterday on my discord to show with forge and comfyui ...will do some more live today
this guy is not good at giving the best settings it seems, and its suspiciously difficult to replicate his samples. experiment for yourself with all possible samplers and parameters until you get what you want.
@captainearlgreyhot449 If you find it suspicious come over discord, I do live and generate live in front of people, I'm no cheater, and work hard enoug to build all those model, I understand your frustration getting result, it's not easy with hyper-sd, but once you dompt the beast, it's crazy good...
me to, using Easy Diffusion, tried the gonzales version someone made as well. And each one looks good until you zoom in and the image is badly blurry. The gonzale's one did nice pharmacy bottles but can't use it this soft. I don't know what was mixed into this model but its soft.
Actually preparing a version 21.55xxx to solve all issue, hope it will be the case..
@mskiller51 Hope so. I tried the art version and it does the same thing. Tried vers below 19 and I get some kind of weight error. Its weird how sometimes models do this. Where it sharp then its like I need glasses. What do you think the problem was?
@frankmike not sure, I mostly use comfyui, even share my workflows and don't get this issues...I am not a python specialist and there is something different with fooocus, a1111 than comfyui, forge is close to comfyui so less issue, I also read your bad reviews and feel disappointed anyway sure it's nit first model that came out...and it's hard to make this one work properly but my samples came all from it, you can download all of them and reproduce it in comfyui. So the models did not deserve that bad of a reviews. In my opinion. I am willing to help but some of you needs to work a little harder before judging or complaining.
@mskiller51 Its not that I want to post a bad review, I don't know how to post the example i'm seeing. But I only use the one program, and other models are quite sharp. I just got something called tessellation and though its a straight lace by the books sort of model its very very sharp. Which models did you use to make it? Maybe I can pin it down on the merge you used. It might have contaminated things. I have something called ultrium - v4 was the last known model to make photos and sharpness. He did something in v5 that did the same thing yours is doing. The most current model he has is now cartoons, i'm wondering if there was too much anime that might be throwing some programs off. Comfy AI confused the heck out of me. I don't use that one.
@frankmike well issue with new nsfw models, is that to perform more sex act merger start to merge pony into the mix which i believe cause that issue and is corrected using pony prompting.
But my first real good nsfw models (the 19 versions) was the first to introduce pony in it, but it not came directly into it, it came from my hentai merge, that at that time did not have that much pony in it but just a little.
Pony prompting seems to make sharper img in 21.5xxx, but I removed the file now....all my samples are legit, but if you check prompt, the best one seems to have pony prompting into it.
Now, I understand Comfyui is scary, then go to forge...you will not regret it
Also, as I send you from chat, I do like constructive criticism as yours, join us in my discord, and you could see all the works we do there... https://discord.gg/BnUhRJW5
@mskiller51 But it does it in your other model too, the art one, or I think all of them. I don't know if there is one common theme that all of them have and that's why my AI thing is allergic to it. I'm curious myself because these models are huge and there is a monthly limit what I can download. I've used a lot of time downloading other models just to run into this now and then. I really don't want want to get into another ai thing. And comfy was just too confusing. I like simple and to the point. I don't know what they left out of that other version and i'm hoping the easy team is working on better models that can work with these other models. But something must have contaminated it.
new version is out, should fix all those issue..
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.