Simile = comparison, likeness, parallel, similar, resembling
Note: Preview images are generated without the use of LORAs
NoobAI ver. [v3.0, v4.0, v5.0, v6.0]
⚠️ATTENTION⚠️
Check 'about this version' tab for more details..
Also, here is the list for the artist tags. There are two files in the zip file, one if the full artist list and the other is the list of artist that I've personally tested to be working and most of them are also from gay artists..
About
To be honest I didn't expect to update this quickly but I can't resist NoobAI's updated datasets. Artist tags also performs much better with less dependent on emphasis to achieve a desired effect. Overall it is an improvement of the already great model, Illustrous XL.
Tips for artist styles / characters
Similar to ILXL, just add in the artist or character tags and that's it. If you're not sure what to write for the tags, just search it on danbooru or here. See sample images for examples on how to prompt the tags..
My Settings
Sampling Method: - Euler a (less detailed and smoother textures)
- DPM++ 2M (more detailed lines and textures)
Sampling Steps: 20+
Upscaler: 4x_NMKD-Siax_200k / Remacri
CFG Scale: 3 - 5
Clip Skip: 2
VAE : Already built-inPositive Prompt:
very awa, masterpiece, best quality, absurdres, perfect quality, detailed background, intricate detailsNegative Prompt:
lowres, worst quality, bad quality, bad anatomy, signature, username, logo, watermark, artist name, bad handsIllustrious ver. [v2.0]
About
Personally I still prefer the default style of the Pony version but the prompt adherence and style flexibility of the Illustrious XL is just too good to pass up. Prompting and tagging is also a lot easier than the pony version from what I've tested. I tried to replicate the look of the Pony version but there's just too much artist styles for me to do anything without messing with the overall quality, oh well..
Tips for artist styles / characters
Illustrious is a fine-tuned model trained on Danbooru datasets, so there are a lot of data on various artist's styles and characters. So you can prompt a specific style or characters without the need of loras as long as they have substantial amount of dataset on Danbooru and are not too recent. See sample images for examples on how to prompt styles..
My Settings
Sampling Method: Euler a, DPM++ 2M / DPM++ 3M SDE
Sampling Steps: 25+
Upscaler: 4x_NMKD-Siax_200k / Remacri
CFG Scale: 3 - 6
Clip Skip: 2
VAE : Already built-inPositive Prompt:
masterpiece, best quality, absurdres, newest, perfect quality, best quality, absolutely eye-catching, detailed background, intricate detailsNegative Prompt:
lowres, worst quality, bad quality, bad anatomy, sketch, jpeg artifacts, signature, watermark, artist name, oldPony ver. [v1.0]
About
Just like with my HomoSimile model, I'm trying to get that specific look that's between 'anime' and 'western art'. I went a bit easy with the LORAs to ensure better compatibility with other styles. So if you want to use a specific artist's styles it wont diminish the results.
My Settings
Sampling Method: DPM++ 2M / DPM++ 3M SDE
Sampling Steps: 30+
Upscaler: 4x_NMKD-Siax_200k / Remacri
CFG Scale: 4 - 6
Clip Skip: 2
VAE : Already built-inPositive Prompt:
score_9, score_8_up, score_7_up, highly detailed, intricate details, detailed backgroundNegative Prompt:
score_6, score_5, score_4, score_3, score_2, score_1, low quality, deformed, bad hands, watermark⚠️DISCLAIMER⚠️
Please do NOT generate illegal content with this model. You are solely responsible for any unethical use of this model(s)
Model License
Description
v5.0
What's new:
Slightly more vibrant colors.
Better response to artist tags.
Slight increase in 'bara-ness' much like in v3.0
Added 6 new artists tags : 'kuromine hiraya', 'ppatta', 'patrick fillion', 'qaisan', 'rennik', and 'tagame gengoroh'. However, the quality and strength of the artist styles might need some emphasis. For example (ppatta:1.2).
Other minor adjustments
FAQ
Comments (133)
Omg it’s getting more and more interesting 🤩🤩🤩🤩🤩
Awesome as f***! Love this.
Here you go, coming with the HEAT yet again! I look fwd to playing with this...😈
Have fun lol..
Cant wait to try it~ My boys~HEEHEE😈😈😈
Hehehe .. 😈
you never miss 👌
Awww, thanks .. :)
I like the fact the 5th version of this model has been released and it only had to wait a few hours to use that version in the image generator on this site.
Yeah luckily I managed to upload it just before the auction was closed so it was a last minute bid that got it onto the generator..
What's changed with v5?
You can check the 'about this version' on the model page..
Kairen92 increased bara-ness is 🍑.
There is only One SD Divinity and Kairen92 is their prophet.
Lmao, I wouldn't go that far .. 🤣🤣🤣
Kairen92 are you gonna update it soonish on TensorArt?
GloberTrotzer Sure, I'll upload it now. Probably gonna take a while tho..
Big Daddy is serving us another delicious version! Thank you, u deserved to be number one on this platform my boy! Make them!
Big daddy? Lmao, first time anyone has ever called me that.. 🤣🤣🤣
Kairen92 Well, then I'll say it, and you deserve that status! Your contribution is simply invaluable and I'm proud of you.
Zignature Thanks lol .. 🙏🏼😅
I'm a bit confused with those two txt files in description. Is the big one has all the artists your checkpoint was trained on?
Kinda, the big one is the full list of artist taken from NoobAI, but most of them either doesn't work (because of insufficient data) or the effect is not noticeable. So the other smaller list is just my personal list of artist that I have tested to be working and its also mostly gay artists. But the list is pretty old and I haven't updated in a while..
Kairen92 Just to clarify. NoobAi itself, as well as your model, both trained on those 35k artists from the big list, and to save time it is best to use those artists from smaller one list, right?
Lewdy_69 Yes, my model used NoobAI as the base, so most models that is trained on NoobAI also have those 35k artists. But the amount of dataset from each artist differs greatly, some has 300+ data to train on, but some only have less than 10. So a lot of this artist tags actually have no effect. That's why I made my own list, just to save time.. You can try your luck with the full list too of course..
Kairen92 Thanks!
At this point, I (mostly) don't bother with other checkpoints. I always get the best results with your models.
I always get excited when I get the notification when there's an update or something new from you. Keep up the good work!
Man, you don't know how much that means to me to hear you say that. Thank you.. 🙏🏼
This checkpoint is unbelievable! I've been having so much fun generating with it
Yaayy .. ☺️
Thanks for this new version.
As an aside, have you considered making a HomoSimile merge based on RouWei? (the epsilon version, not v-pred): https://civitai.com/models/950531?modelVersionId=1832460
It seems to be superior to both Illustrious and NoobAI. I had no concept bleeding and better prompt adherence when testing some of the RouWei finetunes on this site.
There's also a custom CLIP text encoder for it that uses Google's Gemma-3 text encoding instead of SDXL which reduces bleeding further.
A HomoSimile version would go so hard
Hmmm, interesting.. Although I don't know if simply merging is gonna solve the bleeding issue, if anything it'll just transfer the bleeding issue to RouWei.
I haven't tried it yet, so I have no opinion about it being 'superior' to both Illustrious and Noobai. While it does have updated dataset (they mention it was up to April 2025), they also lack any data from e621, which I think is a bit of a loss, especially for NSFW gens. The Gemma-3 version of the model also seems to suffer from inconsistencies, especially when using artist tags.
Tbh, I was waiting for a new version of NoobAI or Illustrious (I know there's Illustrious 2.0 but they basically reused the same dataset, so I don't know if it's worth it) before updating but it doesn't seem to be happening anytime soon. If I decide to make another update, I'll consider RouWei in the future. Although I just updated so it might take a while lol.. 😅
Kairen92 I see, making a HomoSimile version would probably take some time then. But I'm willing to wait 😅
If you are considering an updated ILXL version, then I hope you consider RouWei as a base instead. It's probably only marginally better than Noob but the anatomy/lighting is much better than Illustrious and it has a more up-to-date dataset, even if both ILXL and RouWei don't have e621 images.
And testing out WAI NSFW's finetune of RouWei, I got really excited seeing the model properly split the character's traits in sex images and NOT merging their tags together (like not making half of one character's body dark skin if I had one dark skin character, or knowing not to make a muscular female when there's a muscular male, which are problems that still plague Illustrious and Noob). I usually have to inpaint all those errors away, but using this model felt like a dream lol I guess the only downside (besides lack of e621 data) is the technical floor is quite a bit higher, since it requires properly splitting token chunks
Dears all, hello everyone. I can't upload this model on RUN DIFFUSION + Automatic1111 WebUI; I upload it in the Models folder, but it doesn't appear in the application choices. Can you tell me if there is a way or if it doesn't work on this platform? Thanks in advance for the feedback
I don't know much about Run Diffusion, but on A1111 I believe the folder should be /[SD-directory]/models/Stable-diffusion/
good
Oh a new version? I'm gonna have fun with this one. Always great work, Thank You.
Have fun lol.. And as always, thanks for the buzz!! ❤️
I love this version, amazing colors. I need to test it.
Is this version the same as the previous beta version? Or it is a new one.
Same version..
Don't be fooled, this is one of the BEST models you can use. Not just for guys. it does women incredibly well
Awww thanks! ❤️
I was casually just viewing this page and did not realize that there was a new version released, anyway. IT LOOKS GREAT, and made my men chunkier and huggable. 😭🙏🙏🙏👌👌👌
I'm glad you like it! ❤️
Best model ever! Thank you so much.
☺️☺️☺️
非常好用啊
Does HomoSimile v5 use NoobAI Epsilon 1.1?
Yes. So it's compatible with most Illustrious loras..
Hopefully this checkpoint lasts forever, like literally till the end of the world lol 😘💓🤗😚
I'll do what I can lol.. 🤣
What scheduler do you usually use with euler/DPM++ 2M?
Well it depends, if you're using Automatic1111 webui, I'd just use the 'auto' option, but in Comfy I heard either Karras or Sgm uniform give the best result. I don't really use Comfy so I can't be too sure about that. So I guess just try it out yourself and see which option is the best?
Hello, is “tageme gengoroh” the correct way to prompt for Tagame’s style? I ask because the tag appears to be misspelled in the change log.
Whoops! Didn't notice that. Yes the correct spelling is 'tagame gengoroh'. Just like in the reference image on the model page. Thanks for letting me know..
This model delivers impressive quality even at default —absolutely loving it!😍
Quick question: I’d liked to working with two characters, A and B. Is there a way to control who is topping? I keep getting A topping B, but I’d like to have B topping A, regardless of the order in which the tags are used. Beside, sometimes it mix up two characteristics. Any tips on how to prompt that more reliably?
Hi, there. Thank you for your comment.
It can get a bit tricky working with two or more characters. What I usually do is structure my prompt like this:
"2boys, <prompts that describe both characters, what they're doing, etc>
BREAK
<prompts that describes char A like physical appearance, clothing, etc>
BREAK
<prompts that describes char B>
BREAK
<prompts for other stuff like setting, background, objects, camera angle, lighting, etc>"
As for getting A to top B or B to top A, unfortunately it's just a matter of luck. With hetero couples AI can differentiate them easily like 'male on female' or 'female on male', but with gay couples since they're both male, it can't really differentiate who's who. The 'BREAK' tag can be helpful but it's not perfect, some concept leaking will still occur like mixed characteristics. Hopefully you can get it right the first time.
Also, try adding the tag 'male penetrated' to the character prompt that you want to bottom.
Good luck!!
There is a way of online:
2boy/2girl/1boy,1girl,<what they're doing,background>,<BREAK (if u need)>,
1boy\(<describe the characters>\),<some supplementary prompts>,<BREAK (if u need)>,
1boy\(<describe the characters>\),<some supplementary prompts>,<BREAK (if u need)>,
...
<LORA>
u can refer to the pictures on my home page,it did a good work
In comfy - each of these BREAK sections can be a separate text encoder joined with Condition Concat. This is what comfy does under the hood anyway.
Then, if you have character loras, you can link the encoder for each character through that Lora only. This will greatly reduce bleeding because each character will only have their specific text encoders keywords.
You'll obviously need the model to go through them all, but the clips don't need to.
why is the version 3 not avaliable?
Sorry, between this and my other models it takes a lot of buzz to get them on the site generator so it doesn't make sense to maintain multiple version of the same model.. 🙏🏼
Kairen92 ooooh ok makes sense, thank u!
hello ive been a user of your models for a while now and ive been loving them but is it possible to atleast have one other model say version 4 or 3 available(i totally understand and have seen your comment about it costing a lot of buzz) but the generations with version 5 arent generating images properly for me even with the exact same tags that i used previously, there are distortions and the quality shift to a more "realistic?" is also something.please dont take this message otherwise ,i just want to make a suggestion.
Hi there, sorry you feel that way. But like I said before, it doesn't make sense to have multiple version of the same model on the site generator and not just because of the buzz. It's also because I'd be taking away other people's spot if they want to have their model on the site generator (since Civit only allow 200 models for the generator). You and others can personally place a bid for the model yourself if you like though..
As for the quality of the v5.0 model, I'm not sure what you meant by 'distortion'. I have tested is with different artist styles and tags and it seems to be working just fine. Are you sure you're not using multiple unnecessary quality loras with high weights? Because that's what I noticed a lot of people with distorted image did. Without an example image and prompts I can't really tell 100% of what went wrong though..
quick question: what prompts to use when i need the result to be like a comic page with mutiple panels? i tried "2 panel comic" or "comic panels" but not really working, or, it impossible with this model?
The tags for multiple comic panels are '2koma' for 2 panels, '3koma' for 3 panels, '4koma', '5koma', etc..
For 4koma, you can also use 'square 4koma' for the 2x2 panel version..
begging for version 3 to be made available , probably speaking for a lot of us who use your checkpoints and genuinely love it but there's nothing quite like version 3(comparing to various other models on this site) it gets the job done really good and i wanna use a lot more of it , please make it available for generation 🙏🙏🙏
Sigh, I'll see what I can do..
If you guys want v3 to be usable bid gold buzz on it it needs at least 1k to be eligible for use and please do it, personally it is my fav since it works with flat coloring better compared to never versions and I dont have enough gold on my own to get it up unfortunately.
Oh not again… lets try to contribute to bidding v3, it is my all time favourite and the model i use to work with, i have contributed some of my buzzes and i am really hoping v3 will be available again
Love this current version 5.0! =3
Love this. Any tips for my prompt for AI art, giving a guy balding hair (hair with bald top - a receding hair), I tried it but he ends up completely bald , it may be messy if he is completely bald. But I made some of the guy with balding hair I created as my masterpiece.
I may keep trying. Any tips for receding hair (hair with bald top/balding hair) of a masculine male character in this prompt?
Have you tried using 'balding'? The effects may vary though since on danbooru even slightly receding hairlines counts as balding. You can also use 'receding hairline' if that's the specific look you're going for..
V4 is back??? Wohhooo!
Due to the number of request I've been getting regarding v3, like a LOT of requests (seriously some of y'all need to chill with the comments and DMs, I don't mind getting them but please be respectful), I will attempt to maintain both v5 AND v3 for the site generator. It will probably be available after the next auction on Friday next week.
Thank you for your understanding .. 🙏🏼
I prefer V4 but then again I've never tried V3 on this site, just 4 & 5. Maybe it'll be just as good
I love these great checkpoints too but let's remember that on Civitai only the top 200 highest bids can be featured each week. You must bid on a model with gold buzz for it to be available for generation, which does cost money. It really sucks, and I really do abhor this method with my whole being, but it's the only way they can be used. My favorite is V4, but however I realize it's easier to maintain only the newest or most used checkpoints. However, if you really do like a checkpoint and want it to be available for use, you must contribute gold buzz to the bidding. <3
EDIT2: Disregard, this seems to work great with
ipadapters now. Something must have changed in today's versions. It's fantasatic.
(prior comment)This model is scuffed for any sort of ipadapter work. Not sure why. other models work fine (Including homofidelis)
Edit, seems to be a limit of NoobAI. using a clip vision patch helped https://civitai.com/models/1233692/style-ipadapter-for-noobai-xl
FYI: FTM/Transman/cuntboy content is a lot more difficult to achieve on 5.0 than older versions. The easiest seems to be with 3.0; however, you'll need to tinker with the negative prompt
Hyper content and facial expressions seems easier with 5.0 imo
sad that civit did this, i was looking forward to trying v5. it produces some of the best looking art of the versions but only v4 is even up right now </3
Guys, ver 3.0 is back! Thank you to the people who supported this model in the bidding
Love this checkpoint, one of the best there is! I wanted to ask if there is a way to get curved genitals? Would like to make some of different shapes but all I get are regular straight ones and I've tried a lot of different prompts but I'm not getting the result I would like.
I'd say it's Probably tough without inpainting or a LoRA unless the dataset had very significant amount of curved penises. It's hard enough to try to get the model to understand cut/uncut | flaccid/erect | upside-down/right-side up to all be different variations of the same thing.
That's not to say it can't generate curvature, I've definitely gotten some curves before, but hard to call on it with prompting.
There is a 'curved penis' tag, but given how limited the dataset is on it (only less than 100 images on curved penises), I don't know how effective the tag is. I think it's better to use lora for stuff like this, you'll get more consistent results that way..
Very good checkpoint
This is a damn great model, I love the V4 of it specifically. I mainly use it for humans/humanoids that lean more towards being humans, as it's not that good at adding enough "fuzz" to furries and monster girls. It needs the least amount of retries compared to other models I've used from Noob and Illustrious. The baked-in styles are fantastic, and they work well with most Loras, even ones that are made for Illustrious and Pony. Obviously, this model does men really well, but it does an equally great job with women as well.
Now time for the flaws, the 3 NoobAI versions have very few of them. V3 is pretty good with normal human proportions, but struggles with shortstacks more than 4 or 5. Since one of my OCs is a dwarf, sometimes her legs are too long, other times it's the arms + the oversized breast size than what I specified. (Although I did put medium-large). V4 improves on it by a mile; it's far more consistent with just about any prompts, the only minor flaw is that giving a shortstack medium breasts causes them to have a skinnier frame. (Easily fixed by increasing the importance of "shortstack" or other prompts). V5 continues to make the prompts more consistent and even improves on the lighting that come from windows since it shows what's outside of them rather than just showing some white light more often. The flaw for V5 would be its new default coloring/shading. From what I can tell, it makes nearly anything on the subject glossier than the previous 2 models and has a yellowish hue to everything in general. Took me a while, but the simplest fix I could figure out was to add glossy in the negatives and specify "anime coloring, colorful" near the artist tags to mimic the previous models' coloring/shading.
But seriously, this is a 10/10 model, highly recommend it. I wonder how a V6 could top these 3 even further.
You've done me real good.
Oh no, not again
I feel you, bud
Sigh, yeah.. Civitai changed their rules again. Now they limit the models on the generator from 200 to 100. I know I said I would try to maintain both v3 and v5 but I think I can only do one of them. Please let me know which version do you guys want me to bid on for next time..
v5 seems the most promising and it generated some of the best looking so far so itd be nice to try it, also if theres other sites that offer this model for generation id like to know
Civitai is really pushing my level of patience with it after this rule change. Anyways, I would suggest v3, since it seems more people like using it. I would also like to say, my personal favorite, v4 for the consistency and good generations. I digress though; I'll go with anything in the end.
For me, v3 seems most compatible with most (at least to me) additions so far. I have only tried v5 a few times, so my opinion can be taken with a micro pinch of salt. Since v5 is more up to date, it be the better option to maintain because of civitai's rules.
I vote v5
v5
I vote v5
From 200 to 100 is insane what, that genuinely sucks because I know many checkpoints where many people use multiple versions of it for different types of usage that they found to be comfortable and make the best results with, checkpoint versions aren't all black and white, many times, as we are human, some people who make new versions of checkpoints may not be as good as the other version at some things that some users may prefer.
Also personally I prefer v3 up to this day (tho I do switch between 5 and 3 on cases)
v3
i vote v3
v4 is best in terms of output
I am ok with either v4 or v3. Ever since v5 and v3 have been unavailable, i feel like v4 is another alternate for me to do my work, and i did like it a lot. So please just maintain v4 as much or maybe v3 if you can. No rushing
Or maybe it would be fair to alternate models one per week
@Lewdy_69 it's strange and inhumane, a person has a personal life and also a job, in another he gives free content for which he does not receive money. I would not like to cause inconvenience to the author of the model
version 3 or 4 work best so either of those please
@Amur_X Author wrote himself "let me know which version do you guys want me to bid on for next time", didn't he? I only suggest sustaining one model per week by turns. A fair option for everybody. Given he sustained three per week before, I don't see anything strange and inhumane. If you do, please let me know what exactly is strange and inhumane.
definitely version 3 or 4 since theyve given some amazing results for me
I also want to remind people that you can submit gold buzz to vote for your favorite model. I'm sure your contribution would also ease some of the burden off the model creator. If you feel your favorite model should be available, you can bid on it for it to be featured. With enough effort and teamwork, multiple models can be up at once. If you can bid, remember it's an option. <3
I'd say V4 or V5, V3 is good but when comparing it to the next 2 it's just ok.
V5 is the best, please enable it again soon!
So yeah, since the majority of you guys were saying either v5 or v3, in the end I was like fuck it. I'll just do both for now. But I have to stop my bidding for HomoVeritas XL, so apologies to the people using that one. I don't have any plans for an update, since I'm kinda semi-retired from this site. So for now, I'll just try to maintain the bidding on this one.
Also, thank you to all the kind people who donated me their buzz for the bidding. Just a gentle reminder, YOU can bid for the model you want to use as well, you don't need to donate it directly to me. While I do have an auto-bid system in place, sometimes it might not be enough, especially with bidding becoming more competitive now, so you guys can cover for that in case I'm not here..
Hopefully one day Civitai will ditch the bidding system altogether..
It seems like Civit are slowly crawling towards pay-to-use policy with these kinds of limitations and rules. Shame. I really want to be wrong...
And what destiny awaits v4 btw?
I see Version 5 is available again for on site gen!
BARA BROS! WE ARE SO BACK! 💚 💪
@Lewdy_69 Seems like it'll just be a 50/50 chance on if it can get in the top 100 bids in time. (I mean it seems to have made it at the last second each time since the days before I would never see in there at all. And yet it still somehow makes it)
This model's great.
:3
Is there a way to check which licensed characters are supported, or is there a list?
In the model's description there is two links to danbooru artist styles and character tabs. If you click there you will be taken to a website where all you need to do is enter the name of the character you want to see you and there is a about a 95% chance that character is supported in the model. If that character isn't there, you can always search for a model of them on Civitai. I hope this helps.
I wish this whole 100 spots only thing is reverted back to 200 spots soon, it's just such a limiting number it's sad to see useful versions for anything anyone needs be off for a whole week (I def wasn't using 4.0 for a second who said that, heavy wind today huh)
It will never and it will only get worse
Noooo v4 my beloved!!!
3.0 goty
“I used to pray for times like this”
V4 is finally back (peak, imo)
Hi @Kairen92! I see that you use BREAK keyword in your prompts, but I don't get how you manage to make it work to control your compositions. I see no explanation in your ComfyUI workflow that is embed in your images -- no custom node. Can you give us light on this mystery? Many thanks for all your amazing work!
Hi there, I'm actually using A1111 Reforge, not Comfy. I dunno if BREAK works in Comfy without custom nodes or not but tbh it's not THAT necessary. I'm only using it because I like to separate the prompts to make it easier to understand. For example:
<character prompts>
BREAK
<background prompts>
BREAK
<composition and style prompts>
@Kairen92 Oh thanks! It's interesting that even though you're working on A1111, there still appears to be ComfyUI workflow metadata embedded in the PNGs I right-click save. That's why I was confused. https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Features#break-keyword.
lol @BaraMysteryBox
Hey fam can anyone find a solution to generate tongue that licks the MC? I tried with "disembodied tongue" just like "disembodied hands" but it doesn't work at all.
HAs anyone else been having issues with posting images? I keep trying to post creations but the site isn't letting most if I use any version. Works when I switch to illustrious checkpoints,
The bidding has ended? Anyone?
i think it went to auction around maintenance unfortunately, so now i gotta somehow make gen's from v5 work in v4 _-_
Yes,we can only rebid this week. The deadline for submission is in 6 days and 21 hours.
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.