BeenYou - R13 is up!
For petite lovers
Improving NSFW or LoRA compatibility is not a goal
I'm testing the following versions on BeenYou Lite
⬇Please read the information below🙏
1️⃣Recommended Settings
- VAE is included
- Clip skip: 2
- CFG scale: 8- Sampler: DPM++ SDE Karras
Sampling Steps : 30+- Upscaler (Hires. fix): R-ESRGAN 4x+ Anime6B
Hires steps: 14
Denoising strength: 0.35
Upscale by: 1.5+- Default prompts
Pos: (best quality, masterpiece), ...
Neg: (worst quality, low quality), ...2️⃣Using Adetailer
- Main UI > Extensions > Install from URL
This fixes problems with the character's eyes and face in most situations
The default usage is to simply enable it and leave it alone✅Now you're good to go!
For more information, please continue below
ℹ️Why is my image different from yours?
I can't give you any technical help, sorry
Keep your prompts simple and correct
Tag auto complete extension to find the appropriate tags
Tags like 'Realistic, 8k, ...' sometimes ruin an image
Also These tags are often more useless than you might think
Neg Tags like 'mutated arms' are useless than you might think
Neg embedding can have a significant impact on the characteristics of a model
These are not as perfect as you might think
💝Contact & Support
⚠️Warning
You are solely responsible for any legal liability resulting from unethical use of this model
Don't use my model as bait to drive people to your paid generation service without permission
For anything other than general personal use, please be sure to contact me
Description
FAQ
Comments (148)
Great model ! Is the fp16 version also pruned ?
What models were merged to create this?
Included my personal Lora, very little Sam yang, and Jk stuff.
The base model is not here, and sharing the original is forbidden. (sorry)
这个模型很棒，现在对巨乳大屁股粗大腿完全无感，希望这个模型一直保留。
Wow, love the style!
Very pretty girls, especially if you (like me) prefer tits on the smaller side. Model responds well to SFW stuff, like colour of dress, eyes, hair. Not as great with NSFW stuff. Horrible with loras (at least the ones I tend to use). Slightly better with embeddings, but still far from great. Conclusions: Use it for the pretty petite girls, as they are. Don’t try to do other things with it.
Hey it would be nice to set a version with included VAE
As of Beta 2, Vae is built in
Was this removed? It says no file for DL... I was really hoping to be able to use this for some petite skinny women...
No, the file exists for now. For some reason, my model images and files are gradually disappearing from civitai.
빈유ㅋㅌㅋㅌㅋㅋㅋ
You can pronounce "Been you" to "빈유" in korea.
It means "super flat chest"ㅋㅋㅋㅌㅋ
"빈" is meaning "Empty".
"유" is meaning "Breasts".
sorry sir, it's funny only for me.
i can get it,love form CN
Brilliant. My guilty pleasure.
There is typo in "Prompts: (best quality, masterpiece1.2)" copy pasting it will give strange results. 😖Took me while to figure what is doing it 👀
I didn't understand what you were talking about until just now. I can't believe I was so stupid. Fixed the post, thanks.
I even kept putting in that wrong prompt and pulling images...
@ARIHAN Hehe 🙂
@ARIHAN 🤣 you were lucky to get good images with that, i got horrendous stuff with the typo.
Is this just skinny bodies and tiny breasts or is it the face also. And what drawing style is this? It takes 1 day to download a 2GB file for me so I cant just test whatever whenever especially if i am already downloading another file for 1 to 2 days.
I didn't copy anyone's style. I personally didn't like the excessive body and face expressions and wanted to create my own beautiful models that I liked (with small breasts, of course 😏) .
My model will show the same results as the example image as long as the conditions I wrote down are met. I didn't use any Lora or embedding (which also means it's less compatible).
Tomorrow I will release the official numbered version beyond beta2 and upload detailed comparison images of each version.
Of course, it will get better with each version, but it's more about perfecting my taste. Please wait a day if possible, thanks for your interest.
The style of this model is great! I love it!
i looooove the look of this model .... i hope you can widen its scope over time (males, animals/monsters, etc)
That's my next plan, but it's going to be a while
@ARIHAN awesome :) I ran it through a bunch of tests today and it definitely fell down in a few areas, racial diversity as well as gender diversity and 'fantasy' in general, though it does some wicked vehicles and female cyborgs
There's a typo in the prompt for ALL of my images 🤣
(best quality, masterpiece1.2)
(best quality, masterpiece:1.2)
please don't make the same mistake I did.
Thanks for capturing the essence of my kind woman, not everyone wants Thicc and top heavy women! :)
Definitely love this model! Personally I love the Alpha 2 version best, as it gives more variants when mixed with other face or art-style LORAs. Is there any chance that we can have some other updates based on that version?
Beta 2 & v1 is also amazing, though it indeed doesn't work very well with other LORAs. 😢
This is one of the reasons I didn't delete the old versions.
More update = the closer it gets to my taste, so the less compatible it becomes.
Thanks for liking Alpha 2, even if it's not enough.
I have plans for a more versatile, universal, but still "small-breasted" model.
@ARIHAN Yeah I got your point. I am not opposed to "small-breasted", tbh I love the model's art style, also skinny and small-breasted characters. The reason I love Alpha 2 more is that I can use LORAs to generate different faces, as I feel v1 always results in a relatively fixed looking. But I fully understand it and fully respect it.🙌
Can't wait to see more updates on your model!
Really love your work, like you said yourself, one big issue is how difficult it is to force it to use loras, so maybe that could be improved down the line. Good luck compadre :)
Ai is now the realm of art.
It was so nice to see a new creation. Thank you for sharing your work.
Natalie Portman + chilloutmix(or something similar)?
What happened to BeenYou Lite?
It's down for now because I have some more polishing to do.
@ARIHAN Hey Arihan, I noticed in Lite, there is no color fall-off. I imagine you had some VAE baked in. Would it be much trouble to ask said for no VAE being included? (the washed out look is fine). Also do you have recommendations of which VAE to use? Leaving a comment on the description page would suffice.
@MoreAI I personally use 840000 and have never felt a shortage on any of my test models. VAEs are not that important to me as long as they are adequate (maybe I'm just lucky so far).
I'll consider your request, thanks.
@ARIHAN Sounds good thanks for the clarification, but just to be clear, all I'm hoping is to see future models -not- have VAE's baked in if that's possible. The non-Lite models seem to not have any baked in, so hopefully republished Lite doesn't either.
You know what's cute about this model, and I haven't tested it yet - you COULD use it for some Gender Diversity! Literally :P plenty of Non Binary and Trans people could be represented in outputs from this model <3
Small breasts & Skinny fever
another update!!!!!!!!
Back with R4, with some big changes
You may not like it, but it's my taste, so please understand
I'm running out of creativity and don't know what prompts to put in...🤕
Try it out for yourself and let me know what you find lacking
I'm aware of the problem of picking clothes that are too monochromatic if you don't include clothing colors
I with there would be more models with that skinny look. Recently nearly all Models have changed to generate Elephant bottom parts..
A model with a more realistic style like orangemix / grapefruit would be amazing if keeping that skinny style. Its really hard to fix those messed up body proportions sometimes.
@jackoolado I'm tuning my model so that it doesn't go beyond a certain realism. I don't know exactly what the models you are referring to are, but you can try my experimental Lora, Squeezer.
What was the problem with the previous versions? They seemed pretty good. Never seen someone remove their old versions before :P Some licensing thing?
In short, you're right 😉
Special style, but really hard to control the result, more like a LORA.
Why with the same parameters, the pictures I generated are not sharp enough for yours, especially for the eyes. None or 840000 for VAE settings, neither improves. Ask God to show the way
All you need to do is enhance the face with i2i inpaint. ddetailer automates this process for you. If your environment doesn't allow you to use extensions, you have no choice but to do it yourself.
It's very effective on all models, and a surprising number of people don't realize it. Of course, ddetailer in its default state is virtually abandoned and has its drawbacks, such as ignoring the intended gaze direction.
@Bradcatt I want to understand why the same parameters, the ones I generate are different from yours, is it the reason for my graphics card? My graphics card is GTX 1060 6G ,
@tansinx657 Probably because I use xformers. Additionally, your VGA is a bit underpowered to actively use Hiresfix.
@Bradcatt Just because of this problem, I also deliberately downgraded my xformers to torch: 1.13.1+cu117 • xformers: 0.0.16, the previous one was touch:2.0, but the effect has not changed.
So I am also guessing whether my video memory is too low to cause this, and I can only hope that someone with the same configuration will produce a picture to verify it.
Thank you for your reply.
@tansinx657 xformers itself creates some randomness. Even I sometimes produce quite different images from the same seed.
The data I use for BY inevitably causes problems with eye representation. I tried for a long time to fix it, but failed, and even merging with new sources made the problem worse.
I gave up for now and released the simplest version. It's safe to assume this is the barebones model 😂
Additionally, Adetailer, an alternative extension to the infamous Ddetailer. It's easy to install (Just Work) and works for all models, so give it a try.
This is really great.
Thank you for sharing.
I love this style lol
This has to be one of my favourite models.
It works great on its own and works great blended with other models.
Thanks for putting this together :)
I have not had much luck using lora though.
R11 has largely resolved the nagging issues of previous versions
Although eye expression is still an occasional problem 😭
But how about the beenyoulite?😭
@nbbxs22809 just uploaded an update
I am also having problems with hands. And make the model look at anything else than the camera.
What's the difference between the R11 and the Lite11 versions ?
I think lite gave the model more "curves"
still possible to get the beenyou_a2_a ? (the alpha)
Alpha 2 is here, as you can see
If it's not here, I can't get it either
@Bradcatt it seems that A2 is not exactly the same as the model you used in your preview pictures. The model filename has a difference, the one available is "beenyou_a2" while the one you used for the previews is "beenyou_a2_a". Great work btw, nice to see a model with normal proportions which is not anime style.
@absente1234 probably just VAE baked version
There shouldn't be much difference in the output
@Bradcatt do you by any chance still have it? Or at least the name of the VAE used?
generate some images that fbi will break my door...
Add some keywords to the Neg to avoid running into the FBI :D
@Bradcatt i use ((naked, nude, nipple )) and yet...
Try "clothed" in you positive prompt. or "not naked".
Can't promise it helps. But I have not had any problems like that myself. (yet)
i'll go with things that won't happen for $500 alex.
Keep getting better!
Hands and fingers, poor results
Yes, That seems to be a big weakness with this one.
I also have problems to make the model to look at something else than the viewer / camera.
And it is not happy with use of lora.
But it do make some beautiful images!!!
The hand expression was definitely up to my standards, which is odd.
but I'm a little busy right now, so I'll work on that later
Interesting, for me it made excellent hands and fingers right from the start. Try using the After Detailer, it might help.
Use only the default prompts suggested in the description
Excessive negative keywords and the like will only backfire
Can only agree. And it seems it is locked at "looking at viewer". Tried to make a pic with girl looking at her dog. but have failed so far.
Also I have a lot of problems with strange hands.
Besides that, it can make some really beautiful images.
"Hands behind back" or something like that solves the hands problem ;)
I'm using a lot your checkpoint, the face is always beatifull whit a irresistible smile!
Would you consider to release a "inpaint" version?
(for more info, search: ' how_to_turn_any_model_into_an_inpainting_model ' on reddit)
Thanks for the kind words, but I'm not considering the inpaint version right now due to low demand.
I'll think about it for the next update😉
I'm having this error:
TypeError: argument of type 'NoneType' is not iterable
Time taken: 5.98s
Torch active/reserved: 1825/3256 MiB, Sys VRAM: 5480/8192 MiB (66.89%)
I'm having some issues with your models (wrote on this because this one seems the most problematic). I was drawn to your models because the prompts were simple and the result seemed great. I've copied the gen. data but I'm not getting the same result. Most notably with messed up faces. Any ideas?
Are you using the hires fix? Recently I was having strange corrupped faces at x1.5, upped to x2 and problem solved. Upped to x3 and all faces are masterpieces. SD works bad with low resolution...
@settima_ai are you able to up easily to 3X ? i have a 16gb vram GPU and it runs out of memory if i do anything higher than 1920x1080 (that is 2.5x) so I did set that as a limit for hires fix. It obv. can be upscaled with controlnet easily with the block technique but that's another story, what settings do you use and which upscaler to get to 3X ?
@zx96 It's just because i'm using google colab (i dont have any kind of GPU!) so i'm using their online gpu. I can also access to superior GPU with 50 Gb, if necessary. Powerfull but uncomfortable (slow to setup and prone to internet disconnection)
@settima_ai I've used it, but I'll try to play around with the x to see if I get it. thanks for the reply.
My favorite model, I hope new versions come out.^^
can you make the mouth a bit more, uhh smaller? sometimes the mouth got too big and creepy
Just like my second image 🥲
BY relies heavily on old and poorly trained data, so it's always going to be a bit unstable
A completely new version will take some time.
@Bradcatt yes, thank you! really appreciate your works!
creepy? the preview images are pretty good
@RileyX preview is good for sure, but if you try to messing with the face a bit by using specific prompt, the mouth can be unproportionally large for the face
@Nakoni okay I dont mind that, but after testing I felt like that model has too much glow and lack of details.
Hi!
Beautiful model and I really like using it. Out of all your models, this is probably my favorite. Though, I need to ask.
I'm writing a family drama story with some lemon elements, and I was wondering if you could add a bit more NSFW to this model, or create a new checkpoint with the same style but more for NSFW purposes. The artstyle is absolutely my favorite and no other model provides such quality as this. Sadly it doesn't like LORAs so that's quite limiting.
It's kind of funny for me to say this, but I'm not interested in hard NSFW at the moment. I'm working on the next version, but BY is always focused on the body shape itself.
I have a few NSFW LoRAs for BY only that I plan to improve, integrate, and upload though.
@Bradcatt It's your model and you know what you want with it, so I'm fine totally with your decision. Though, if you ever change your mind, I will be the happiest person on the planet. Creating NSFW images (not necessarily orangemix level of NSFW) with perfectly proportional beautiful women and such lovely style would be a dream. I would try merge your model with something but I would probably just mess everything up, I have no experiece at all at merging models.
I am looking forward to your NSFW LoRAs and the next version of BY.
can you please reupload been real? it went off.
Maybe better avoid some trigger words
That's not my model.
@Bradcatt got it. sorry
Uploaded an unfinished version of R12
My SSD crashed and I lost data, so I'm recovering for now
uploaded file is rv15, and the example images are mostly rv14, which is not reproducible.
It's a bit unstable, but it works.
I guess it was a mistake to put all my data on an external SSD 🥲
I solved this problem years ago by having TWO exernal disk and copying always data on both. Of course, it may costs some $ more but it pays back a lot...
wow... thats really rare to hear an SSD dying like that. You tryied dismount the enclosure and mount it internally?
@p_p To be precise, it's an NVME, but I used a cheap enclosure and ended up with data contamination.
I could upload the models I have saved on hugging if you desire :)
WOW R12 seems promising! I like this path! By the way I have also tested BeenReal (a merge of BeenYou and a more realistic 2.5D model) and it is fantastic for both super cute faces and slick bodies, you could try to reintegrate it lol!!!
Hi @L0L1 , Civitai banned to STYM, the BeenReal checkpoint creator. Can you upload the checkpoint in MEGA or Drive and share link please? Can you write me in Twitter? (@brekuji)
Yeah interested in that too!
Yeah, as soon as I receive the notification of new update it was removed from civitai
What was the problem that made it disappear?
Hi, can you or smb send it to me? @pipiskalka (telegram / twitter)
need too!
the user STYM uploaded tons of models I think, he could have got reported for some reason. Honestly, civitai happens to ban people without notice for no apparent reason. I guess their banning system is completely arbitrary.
@Brekuji @meproimmocroye6500929 @Cala_Cunca @Hrenovina BeenReal is one of my favorite models recently, it has great faces most of the time without the "same face" syndrome.
Here it is : https://xgf.nu/XCFhR
@L0L1 Thank you very much! I'm gonna test it!
By the way, if I remember well, BeenReal was a merge of BeenYou and midreal25DAnime (which has been deleted as well it seems... but I can provide it if you want)
Well, so here is midreal25DAnime_V2.1 for those who want to try all kinds of merges!
If someone has a hugging face repository and uploads them there tell me!!!
https://xgf.nu/UdEBw
@L0L1 I merged beenyou (lite too) with revanimated and it's realy good too.
@L0L1 THANKS!
hello everyone I have asked in the Discord official channel about STYM and they told me he was banned as he violated the TOS of the site, but when I asked for further information, I didn't get any answers. I see that the author of BeenYou was ok with merges so it wasn't that problem. If you decide to create a repository of his models, I have a few archived. The best would be to get in touch with him, but where is he?
@zx96 mind sharing the repo please?
@Nakoni there is no repository, i have models on my own hard drive, i was referring that i could help if there is going to be a place to archive all STYM models, nothing is online except what linked in this thread.
@L0L1 I don't think that link works anymore :(
@zx96 if you do end up hosting them somewhere please share here https://www.reddit.com/r/StableDiffusion/comments/11pppvw/keeping_track_of_nuked_loraembeddingmodels_from/
I'm sticking with R11 for now...The new art style is not for me..
Face is close to the default and completely unfinished, but the colors are going to stay this way.
R12 looks promising!
I'm not sure yet if I like the new contrasty look, but I will fiddle with it today and see what I can create. I wonder if this revision will be able to handle LoRAs better, considering this is the completely rebuilt BY.
Looking forward to the finished rev!
And make sure to not store everything on your external SSD next time, ehe.
This is not an entirely new BY
It's the last version that uses the existing dataset
That's what happened, intentionally or unintentionally 🤣
Oh, so I'm guessing the one after this will be the new BY. You did mention on kofi that it will be rebuilt. I'm also guessing that it will take a looooong time. 😞
Referring to R12-unfinish : not the right way!
(I dont have 'adetailer' because never managed to install it, for unknow reasons)
I'm doing some comparative experiments swiching R11 with R12u, same parameters and seed. Both seem to do well with very simple prompts. The problem is when you add more. While R11 remains extremely good, the R12 version produce always bad results whith bodies and faces caricatural, like if subject had a bad plastic surgery! Looking at your demo images i try to guess: you wanted a nose even more raised and more red blush. But this was overcooked and ends in the problems I said. Also while R11 accepts famous people to change faces, R12u will literaly overact producing atrocious results... I could post some comparison but this fact is obvious after few time: R12u produces always worst result, sometime much worst...
I think we need to keep in mind that R12 is still unfinished. The author stated that it's unstable, but works. I would think by the time the stable version comes out, things will get better. But I'm not experienced with these stuff so perhaps I'm wrong.
I agree. I like R11 more because it's subtle and not too saturated.
is it forbidden to merge your model with others? because a user has been banned and he published a model merged with this one, just asking to understand what is the policy, thanks!
There are several other models with BY as a material and they are still available for download. I am not against merging models and the user being banned had nothing to do with me.
@Bradcatt @zx96: are you referring to "BeenReal" model, may be?
@settima_ai yes, cause I wanted to understand what happened there. the user had some great models that did not seem to have anything wrong within, but I was away a few days and I'm back and he's banned, wanted to understand why and if his creations are somewhere else, where. I was assuming there was a problem cause he used this famous model in his merge.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.