Update 19.06.2026. Lustify V9 (Zenith) IS OUT!
This is the absolute final version of Lustify built on the SDXL architecture. You can find the full changelog and a breakdown of the new features there. <----------
Recommended Workflow.
Alternative Access & Support: Outside of CivitAI's Early Access, you can get the model via my Boosty.
Just like last time, after the early access ends, there will be a 1-week period where the model is available for "on-site generation only." After that, it will become available for everyone to download.
Alternatively, you can make a direct Bitcoin (BTC) donation, just make sure to DM me after that: 1CoMZvrXtRXbz7BCzdvxifSjHqhaf1yUZs
LUSTIFY
Lustify is a photoreal checkpoint optimized for creating high-quality imagery of women in explicit/sexual scenarios. The model has been trained on a hybrid dataset, meaning it natively understands both Danbooru tags and natural language prompting.
Note: While it has a strong NSFW bias, if you guide it properly, it is fully capable of generating stunning SFW photography, objects, animals, and fantasy environments.
Crucial Points Before You Download:
NOT Pony or Illustrious-Based: This is a standard SDXL checkpoint. Do not use Pony/Illustrious LoRAs with it; they will not work correctly.
Distant Shots & Anatomy Limitations: Like 99% of SDXL models, distant shots can sometimes suffer from distorted faces or hands. This is an inherent limitation of the base technology (4 channel VAE), not the finetune. It is incredibly easy to fix using Inpainting or ADetailer. Please utilize these tools rather than reporting this as a bug.
Experimental Nature: My training workflow involves a lot of unconventional, high-risk experimentation. Because of this, some prompt combinations that worked perfectly on other checkpoints might behave differently here. If you encounter unexpected behavior, feel free to give feedback so I can look into it.
Recomended parameters:
Sampler: DPM++ 2M SDE/DPM++ 3M SDE,
Scheduler: Exponential/Karras
Steps: 30
Cfg: 2.5-4.5.
Highres.fix: upscale by 1.4-1.5, denoising ~0.4
For the distant shots of people you probably wanna use highres.fix and/or adetailer.
Also, use reasonable positive/negative prompts, shizoprompting does more harm than good (with any SDXL model).
Some tags that have a big visual impact with this model:
Camera type tags: "shot on Polaroid SX-70", "shot on Kodak Funsaver", "shot on GoPro Hero", "shot on Canon EOS 5D", "shot on Leica T".
Photography styles: "analog photo", "glamour photography", "street fashion photography", "candid photo", "amateur photo".
Lighting types: "cinematic lighting", "neon lighting", "soft lighting", "dramatic lighting", "low key lighting", "bright flash photography", "warm golden hour lighting", "radiant god rays".
Film types: "Ilford HP5 Plus", "Lomochrome color film", "Fujicolor Pro".
Photographers: Alessio Albi, Martin Schoeller, Miles Aldridge, Oleg Oprisco, Tim Walker.
Others: "film grain", "bokeh", "dreamy haze", "technicolor", "underexposed", "low quality", "lowres"
Alternatively, you can give a read to this wonderful prompting guide by PromptGeek. It has A LOT of useful tips I didn't know about even after years of using SD.
Description
FAQ
Comments (146)
This is the best NSFW non-pony model I have tried so far. Great job!
Thank you!
It's really hard to find XL models (non-pony) capable of producing truly NSFW content. (apparently the censorship gets heavier with each base model released), but thanks to the community this is resolved. thanks.
can you make it available on rundiffusion
I think it should work now.
its on thinkdiffusion, not run diffusion
@anubhavk198 1) I don't pick what platforms it is available on.
2) It is on rundiffusion. I've even created an account to see if it is. You need to download the model manually inside the service.
@coyotte Thanks!! what a great model !!!!!
@anubhavk198 thank you, glad you're enjoying it
I fell in love with the authenticity and actual realism of BIGASP, and this looses a bit of it but it's worth the cost! Results are much more stable. Thank you for this!!
I see there are some tags in the description but, would it be possible to have a list of new triggers that aren't just visuals or quality? like poses or actions or anything?
This model isn't made to be some kind of relative to BigASP. Actually, as of V2 there's only ~15% of it left inside.
Even as of V1,
The recipe was something like: 0.55*(my unreleased SDXL finetune)+0.45*(0.5*(bigASP)+0.5*(Pyro's NSFW)).
So BigASP was there only by ~22.5%. Pyros was ~22.5% as well, the rest remaining 55% is my personal work.
So, that means comparing Lustify and BigASP doesn't make much sense.
There are no new triggers, only the expansion of the existing ones that you all know and use daily. Some parts of the dataset are captioned with NLP, some with danbooru tags, some with both.
I don't know how to take all the tags from the captions and put them into one file so that you see what was used. And it's not really useful, since in this case the NLP descriptions would be considered as tags and the whole file would be littered with single instances of the NLP descriptions of some particular pictures x100
Thank you for your comment tho. I appreciate every opinion.
this model is amazing
i love especially the vivid color
Thank you!
It's the darndest thing!!
I followed the models instructions and my images got SO MUCH BETTER
whodathunk
Haha that's why I write all this stuff. To help y'all use the model the way I intended it.
what instructions?
@richai
Recomended parameters:
Sampler: DPM++ 2M SDE/DPM++ 3M SDE,
Scheduler: Exponential/Karras
Steps: 30
Cfg: 2.5-7.
Highres.fix: upscale by 1.4-1.5, denoising ~0.4
For the distant shots of people you probably wanna use highres.fix and/or adetailer.
Also, use reasonable positive/negative prompts, shizo prompting does more harm than good (with any SDXL model).
@richai so in short,
NO 2m karras/euler a, NO 20 steps, NO 10 cfg, NO wall of schizophrenic negatives.
amazing model. better results than juggernaut or pony imo. round of applause. 👏👏👏
I'm glad it works that good for you. Thank you!
Although I'm not sure if it's fair to say that's it's better. Jugg/Pony are needed for other purposes that this model isn't as good at.
Thank you so much for putting so much work into this, I appreciate all the work you've put into this! I really love your work, you're by far the best at this type of work in my opinion, and I cannot wait to create some beautiful images with your new version.
Thank you so much, I truly appreciate your kind words
This is right up there with Anteros XXXL, BigASP, Big Lust, and Pony. Thanks.
doesn't work with adetailer
Meaning you do something wrong. It works. I use it all the time.
Sadly the same for me in V2. When AD inpaints theres only a totaly oversaturated mess. Strange thing is it occurs only after a while. If i switch back and forth between V2 and a different checkpoint the other one is working as always. Back again to V2, still broken. Only fix atm is to close Automatic 1111 (not restart GUI) and start fresh. Then its working until it isnt :-(
I've found the same thing with V5. It worked for a day and it worked great. This checkpoint is fantastic! Then it stopped working with ADetailer. I've restarted many times (even restarting the computer), re-downloaded even, and nothing. Every ADetailer turns into a rainbow saturation. No other checkpoint that I've tried does it. It does a decent job on eyes (which is all I really use ADetailer for most of the time) but ADetailer would be great for the other times.
@valdd I've NEVER had something like this happen to me. Make sure every ADetailed setting is set properly (open the ADetailer tab, tick "custom resolution" or something like this, set the resolution to 1024x1024).
@coyotte Yeah I haven't either. I tried increasing the resolution but the effects are still the same. All other settings are the default. Anything within the Adetailer mask becomes a rainbow mess. This doesn't happen on your DMD2 checkpoint though.
played around with V2 for a while now. Really amazing!
is it possible to train a LoRa using this model as the checkpoint? it could yield interesting results
Works really well in foocus
What did you do to get In-Painting working ? For me, it stops at the "preparing task" pre-process.
@JohnnyWu22 try different models, like juggernaut or realistickstockphoto
One thing I realise by testing V1 and V2 is that faces on V1 are more realistic and natural than V2. Maybe is something personal.
Also is true that V1 could be to charged to NSFW, for example putting someone in a kneeling or laying position most of the time there's a man below, no matter what, even if you put negative prompts.
Imo V3 should be the perfect blend between V1 models and V2 versatility.
Your work is something else and you must know it.
Thanks!
A couple of questions -
- How do I get this to work in Focus 2.5.5 ?
When I tried, it stops at "preparing task" and doesn't generate any images. (Focus 'Enhance style' is OFF)
EDIT - It won't even generate an image, not just in-painting, same reasons as above. Normal Lustify v2 and other models are still functioning fine.
- Assuming someone tells me how to solves issue #1, can I delete Lustify v2 (to save SSD space), or do I need BOTH models ?
"How do I get this to work in Focus 2.5.5" what are you referring to? Lightning/inpainting/Normal?
@coyotte SPEED mode selection and Inpainting, I didn't try lLightning. The fact that I can;t get it do anything suggests I have messed up badly somehow. I put it in the models folder with all the other Focus models I downloaded before.
@coyotte Just tried with all 5 of Focus's modes, and they all stop at "preparing task" for me.
@JohnnyWu22 please be more specific. Are you using the regular checkpoint? The inpainting one? The lighting version?
@coyotte I am using the file called lustifySDXLNSFWSFW_v20-inpainting, which I downloaded from the link on this page, which I placed in the Checkpoints folder of Foooocus. I am new to AI stuff, sorry.
As I was typing this, I just thought "Am I NOT supposed to select lustifySDXLNSFWSFW_v20-inpainting as a model in the Foooocus drop down box (which I have been doing), and am supposed to ONLY select the vanilla Lustify (lustifySDXLNSFWSFW_v20), and this then automatically uses the inpainting version ?"
@JohnnyWu22 okay. With focus you don't need to use the inpainting models. The whole point of focus is that it uses it's own engine for inpainting, which turns any model into an inpainting one. So, that's what's causing the error - it was trying to turn an inpainting model into an inpainting model once again. The same applies to lightning. You only need the normal one.
@coyotte So the file I downloaded was NOT meant to be used as a selectable model inside the Fooocus drop down menu ? That particular download is for Lightning ? I am only supposed to select Lustify v2 as the base model, and Lustify v2 (optional) as my refiner ? And then Fooocus will perform the inpainting using that base model and that refiner ?
(Just putting it down in writing incase someone else comes along with the same misunderstanding)
Therefore I should DELETE the file I downloaded ?
@JohnnyWu22 Yes. You only need the regular one for Fooocus.
@coyotte Thanks for your help and your patience.
The model looked good, so I immediately tried it out in the WebUI. While it was being generated, it looked like it would output good colors and composition, but the final product is low resolution and the colors are eerie, like a photographic negative. Is there something wrong?
Probably.
Please provide the settings you are using. Sampler, sheduler, resolution, steps, cfg, ui.
The inpainting version of this checkpoint will not load into A111, the others all work fine. I re-downloaded it twice. Anyone else have this issue or know of a solution?
I do not know it wont load for me in fooocus either
Good for using normal prompt but very bad at using with character loras. You can get the likeness in first generate sometimes but after that it becomes uncanny valley.
I mean, the checkpoint was trained for a long time. The weights have shifted a lot from the base ones, that's why your character loras don't work as good. I won't compromise quality for compatibility. If you like the checkpoint, consider creating the character lora specifically for it.
@coyotte You think it works if i create lora for it? because 1k buzz for a lora training is too much, hope it worth it.
@chrisss1 someone told me they did exactly that and it worked for them, so... I think yes
THis is the best model, I have been using this v2 since day 1 and had no issues in 99% of loras i Forge UI using Adetailer. this is by far the best nsfw model on CIVITai
@neo_neil_71 Hey, did you also tried the new v3 and inpainting in this checkpoint?
@chrisss1 actually i only use the lightning and turbo models model i never use the normal version, so i am not sure about it.
The best checkpoint. A+. If you've ever thought "I wish these Pony realism checkpoints actually looked 'real'" then this is the one. This is the only reliable checkpoint that understands Danbooru tags and looks like real photography. Have tried all the "realism" Pony checkpoints, and they still have that ever-so-slightly fake pony look to them. This is actual real life imagery, but understands the same prompting as Pony.
What's the recommened settings for inpainting model?
Hey there! Great checkpoint!
I'm using your refiner workflow and running into something. My base model is generating anal but no matter what I prompt for with the refiner, it seems to consistently turn anal sex into regular sex. Any tips on avoiding this? thanks
I just wanted to share the setting I found work best for me.
Steps: 12
CFG: 2
Sampler:DDPM
Scheduler: Normal
Most typical Resolutions work well
Positive prompts i like: POV, Portrait or close up in some cases, keeping the subject close to the camera will increase detail.
tnx
Thanks a lot, those settings are pretty darn good!
@FlyingDutchman72 thanks brother
Congrats on an excellent ckpt, been top 3 for me for a while, thanks :)
Thanks a lot, I'm glad it works that good for you!
Hey guys, try out the model with my Lora
Anybody have a recommended inpainting workflow in Drawthings? the recommended settings don't result in anything good. Cheers!
The same ones i use for everything. 2m sde karras/exponential, 30 steps, 4 cfg.
Lustify V3 incoming.
I'm doing some final training passes. ETA: this week
3...2...1........
My new favorite model.
Thank you!
I am using the v2.0 inpainting version in comfyui and an error occurred. The error message is "Given groups=1, weight of size [320, 9, 3, 3], expected input[2, 14, 160, 119] to have 9 channels, but got 14 channels instead", what should I do?
you on comfyUI?
As far as I know, this type of error occurs when a wrong VAE is used.
can you share a complex comfyUI workflow to get better images with your model?
OR you can try this one, if the one above isn't what you're looking for
https://civitai.com/images/23755643 (Press on Workflow: Nodes) to copy the workflow.
Although for all the preview pics I use only AUTOMATIC1111.
GOAT model wow
Thank you for your awesome work. v3 is good but I still prefer the sharpness of v2, even for hardcore.
Either way, these are the best nsfw models out there, by far.
Yeah, I can understand why someone would prefer V2. It's all matter of tastes and use cases ✌️
Thank you!
@coyotte I really like your checkpoints! I tried V3 today and saw some improvements, but no matter what I prompt or what parameters I choose, all the skins are really smooth, which gives a fake and “less amateur” feeling. V2 works better at this (sometimes even too much). Is there a reason for this?
What sampler and scheduler are you using? These have huge impact when it comes to textures.
Also, V2 and V1 were broken with "DPM 2M" sampler, it was giving a really grainy/artifacted image.
For some reason, people kept using this sampler and were posting their results. Neither it looked good, nor it represented the models visuals, so I've tried fixing this problem. It might have resulted in slightly less details overall. Try using highres fix, it should bring back all the details
@coyotte I use ComfyUI and have tried many combinations of samplers (dpmpp_2m, dpmpp_2m_sde, ipndm, ipndm_v, deis, euler_cfg_pp, dpmpp_2m_cfg_pp) and schedulers (exponential, karras, ddim_uniform, beta, ays, gits). After some testing, the best ones so far that give better skin details are dpmpp_2m_sde and dpmpp_2m both with karras. The beta, ays and gits schedulers give extremely smooth skins.
I was thinking, could the reason for this smooth skin be the use of low quality “out of focus” images in the dataset? Because many images I generated gave me the feeling of “ bad focus” of the skin, even though they were actually in focus.
@LG89 i don't think so. There's like 50 "bad pictures" in the dataset of 1700 pics. That's almost nothing
Gotta be something with your settings or prompt, I'm getting way better skin details than V2 without prompting for it
@onionhulu781 Yeah exactly. As I said in another comment, my workflow was built around V2 and I made it to compensate the overly grainy/artifacted images of V2. Now V3 works like a charm!
V2 was hands down the best realistic NSFW checkpoint around. Looking forward to trying V3!
Thank you! Would like to know what you think about V3
@coyotte Hey I'm sorry if I contributed to your stress at all, I deleted my last reply. I just tried to offer helpful feedback about what I thought could be improved. Don't let the criticism get to you. Your models are great and as long as you like them, that's the most important part. Remember, it's always the people who complain that seem the loudest. I guarantee you the majority of people are grateful and appreciate your work. I know I'd be super bummed if you stopped making your models. V2 and V3 are honestly the best realism NSFW models ever.
@nncyberpunk no-no, the feedback in comments was OK and reasonable, you didn't have to delete yours. It's the things that were said elsewhere that made me do what I did
@coyotte yeah but I'm sure my comment didn't help. Just know that people appreciate your work.
I have been using your workflow and checkpoints for all my art, or smut, rather. Your last update is fantastic; your SDXL model renders all your nsfw as_is, great prompt adherence, milk looks fresh from the cow, faeces look good enough to lick. Wait...What, i meant faces...Thank you.
Hey, I'm glad it suits your use case so well ✌️ Thank you for being there
V2 felt more like a side-grade to V1, but V3 is a clear improvement imo! Very natural looking and improved skin textures on a brief first impression. Also better face variety, much easier to get women with less makeup and less "porn" look haha. Awesome work!
V3 is great so far, i'm actually seeing a improvments. Used normally, then i even tried pony loras with it, some interestingly works some not as usual. Will update this after try other things too. My only problem so far, i am struggling to get a anal scenes with this checkpoint lol, most of the time getting vaginal despite prompting anal. No idea why. Thanks for your works!
v2 is my go to for anything and everything NSFW, it just does it all so well.
However v3 doesn't seem to work with in-painting (Fooocus), it just puts dark black clothes, or dark black skin in place of the painted out clothes.
Do you use normal v2 for inpainting or v2 inpainting version?
@chrisss1 I used normal. There is a long post on here between myself and Lustify's creator, where I tried to use the file listed as the inpainting file. Apparenty, it turns out you just use the normal standard v2 of Lustify for inpainting in Fooocus.
When I tried the inpainting version of v2, it wouldn't generate any images at all.
@JohnnyWu22 exactly. As you remember, fooocus uses it's own engine for inpainting, it's own controlnet-like models (you can use them in comfy with specific nodes as well). It doesn't stack with the inpainting model's architecture, hence you couldn't generate anything.
The inpainting checkpoint version is for Forge/A1111, not sure about Comfy
@JohnnyWu22 Thanks, i tried normal v2 for inpainting and it works just good. Not the same with v3 tho :(
@chrisss1 Yeah, I can't get v3 to do anything except add black clothes to women i'm trying to strip naked :-D
I tried it with a variety of different Refiners, but it was still the same.
I CAN use v3 as a Refiner. I don't know if it makes any difference to the output ?
@JohnnyWu22 just use v2. For inpainting it's still pretty good.
@JohnnyWu22 Didn't tried v3 as a Refiner but no idea if it makes difference. As coyotte said, better use v2 for inpainting. Personally i use v2 for inpainting.
Nevermind
I've deleted V3.
V2 is the final now, I won't work on this anymore. I never intended for this model to be as popular as it is now, it was more like a local thing. Because of it's popularity I have to meet the expectations of other people now, which isn't what I want or need in my life.
Thank you to all who loved my work, sorry to all who was disappointed with it.
I understand what you mean and respect your decision.
I am one of those who downloaded v3 and I can guarantee you that I will continue to use it, because despite everything, the improvement can certainly be seen.
Surely the people who criticized it have little experience/patience to be able to "master" a good model and, in my opinion, this is it.
Having said that, I still hope to see more of your projects because so far they have always been excellent.
Thank you
ahh.....but..... anyway, thanks for you work! I love your v2 and v3 and inpainting models!! Good luck in other works! ((masterpiese:100)), (best quality), (thanks:100)
@BobDF thanks a lot! I'm glad you found a way to use it.
I have another unreleased model in a stash. It will be released if/when it looks good enough to me. It's not so NSFW-focused tho
@markibnfadlan447 thank you! (Ultra super mega highres extremely detailed:100)
Ah shucks, I never got round to downloading it! Any chance you've uplaoded it elsewhere or can otherwise share it with me? I'd greatly appreciate it. Thank you!
Don't get me wrong but honestly, i don't get it. V3 was such a good improvment. Why a someone wanted to remove something they put such a work and effort in it. You remove it because you don't want it to be a popular? you remove it because you have to meet the expectations of other people? idk man. You don't have to meet no ones expectations and your model being a popular not going to harm you. Yourself said this in the description of the model:
"All of my stuff is very opinionated. I create models mostly to suit my needs, my workflows, my prompting, etc."
Yet you still saying remove it because you have to meet the expectations of other people? kind of an contradiction if you ask me.
@chrisss1 I removed it because it was the most opinionated piece of work I've uploaded ever. It had some things that worked only for a few people. "Lustify" being popular means I have to put out models that will work for the masses, not for just a couple of enthusiasts.
I got shitted on pretty hard after I released v3. Not in the comments. On the imageboards, in the DMs... Also, this decision was made on emotion - it's entirely possible I'll revisit it at some point. Or maybe I'll just make a less opinionated version. Or maybe I won't... Idk 😒
@coyotte Dude i understand you. People will always find something to bitching about. Forget about them whiners and focus on your works, don't let your emotions get the better of you.. Please do revisit it, this is the best NSFW SDXL checkpoint i have ever used. Also works really well with img2img and inpainting too. It got big potential, evaluate it. Also there is Flux too, consider making Fluxtify, we need you man!
I understand how people unfairly put their expectations on you. It's crazy that they demand things from a free tool. I love your v3 model and hope you release future models.
I came back to give credit on what a fantastic update to this model it is, and then see this. I'm sorry to hear you're getting negativity sent your way. Focus on the praise as there's a lot if it and rightly so. Let the emotions settle and I hope you continue on becasue it's greatly appreeciated. Thanks for the work you put into it all.
@chronosafe @am3ient2 @chrisss1 thank you all, I really appreciate that. You've made me change my mind. I will upload it back
@coyotte The whole point of this AI thing is to make things that YOU enjoy. Every model is going to have its strengths and weakness, quirks and such. Each model is unique (for the most part). V3 is awesome, V2 is awesome and V1 is awesome. The inpainting's are also insanely good. Don't let people have real estate it your head.
@coyotte Great that you will make v3 available again! Please forget about all of these idiots that complain about it. These are usually the people that don't know how to improve generated images and expect all models to output the very best possible quality. Totally unrealistic expectations. You created some of the very best models available. Even if one or the other of your models is a bit less good, we still want to use them.
@coyotte You did a great job with V3! I came here just to tell you! Sorry about my "bad first impression" in the other comment, my workflow was built around V2 and I had to readjust it. During these days I re-evaluated V3 and found how good it is. I can't understand why you have been criticized so much, you made fantastic checkpoints. Thank you for your great work!
@LG89 your impression was completely reasonable. thank you!
@akjskdjfkjs it's back
Brother is having an exitential crisis. Just do what you love and dont listen to others. Simple as that.
Loving the model a lot and curious where it goes. I like models that surprise me :)
Man, your models are the best, really appreciate your work. Looking forward for your new models/workflows
@coyotte the same thing happened yesterday to SG when he uploaded RealFlux 1.0b. Someone in the comments said and a quote: "May i ask you why do you upload if your models eg. 1.0 Dev is currently unstable and many things to fix? Why don't you release after you fix things. Like Qaulity over quantity? just askin".
Like wtf is wrong with you. The guy is uploading content for free and now he is supposed to maintain your bullshit standards? From what I read, doing a Flux checkpoint from the distilled version is a very hard thing to do, and he just published it so it can be there if anyone wanted to use it. If you don't like it, don't use it.
Popularity will bring haters, always. Even if your product is the best, there will be haters. Personally I think V3 is a massive improvement over V1 and 2. Your work is fantastic and it would be a real shame if you stopped due to the loud minority of haters. Most people only comment if they have something negative to say. I say keep up the excellent work and don't let em get to you!
No matter what you do, You will ALWAYS find a annoying group of people that complains about it. So just make the things you love or the people that love it! Image Rock groups stop making Rock Music because Techno lovers don't like Rock ;P
@WhatTheGuy @onionhulu781 @sakadip918 @lannister34 @Robbedoes @binauralhealing100139 thank you all for the kind words and support, I really appreciate that 💪
V2 was the best, V3 is still the best! But I think V3 is slighty overtrained in some areas where V2 had more freedom.... Still love it, can't wait for V4 =D
Thank you very much for reuploading v3, you work is amazing!
About this
V3 - Combines V1's porn capabilites with V2's aesthetics. The visuals are darker than the other two versions (can be countered with prompting if you don't like the darker colors).
How would I go about that? I tried using prompts like "studio light", "perfect light" and such but nothing seems to change. Sorry if this sounds obvious but I'm still a newbie with SD
Make sure you are using VAE (example sdxl.vae.safetensors) if the image is too dark
@arkaymail VAE should be built-in in this like it is for V2 (I assume?) - Its mostly that exposure or something is a bit low for every gen. Images still turn out good, just a tad darker
@tiosolid you can use "underexposed, high contrast" in your negative prompt. Most of the time it's enough
@coyotte Thank you for the tip! will try it!
This is quickly becoming my favorite model, thank you!
thank you for using it!
I really love your work, I've tried the newest V3 and it's my current favourite rn. If it's possible can you release the inpainting version?
I will release one for V4 (coming out soon)
I don't comment much, but dang!, this is the best realistic model for SDXL, Since I can't run Flux this is Amazing. Props to you.
v3 is probably the best when you want to make some dirty images. ;)