Welcome to Eros, the temple of human nudity.
The goal of this model is to reproduce human nudity with high accuracy and controllability on various body types, ethnicities, styles, etc without shifting the base model too much.
Eros Models Plan:
Eros XX (Woman nudity) - Released, version 07 - ONLY photographs for now - next version will include anime, cartoons and others
Eros XY (Man nudity) - Unreleased, if you are up to help enlarge my dataset, PLEASE contact me
Eros XXY (Human nudity) - Unreleased
Recommended Settings:
FLUX KLEIN:
I strongly recommend to use distilled Klein9B and not Klein9B-base unless you know what you're doing.
FLUX KLEIN edit capabilities are enhanced by this Lora, the knowledge has been for the most part well learned so it can transpose the concepts properly. (most of the time)
For distilled
steps: 4-10, sometimes more for complex poses.
CFG: 1-1.5
For base (don't!)
steps 20+
CFG: 4-8
Samplers: Euler/Simple-Beta works fine, I didn't had the time to compare much for now
FLUX DEV:
steps 20-40
CFG 2.5-5
Samplers: Euler/Simple or Beta, DDEIS/DDIM, Flux realistic
All were done on forge webui (thanks lllyasviel)
Eros XX:
/!\ Remember weird pose = more steps /!\
Here is a list of what is supported right now by the model with the following self rating :
Good (almost 100% consistent), Nice (~80%), Meh (~50% and/or need precise prompt), Hu (try your luck), unknown
Poses: standing, sitting, squatting, resting, lying, crouching, slumped, slouched, on all fours
Camera angles: selfie, from behind, from the side/profile, top view, rear view, high angle/low angle, from below/underneath
Close-up: chest, lower body, pussy, breasts, buttocks, feets, hands
Breasts sizes: tiny breasts, very small breasts, small breasts, medium breasts, average breasts, large breasts, very large breasts, huge, enormous
Aerolas: small, large
Trigger words: topless, nude, naked
Ethnicity: Caucasian, Asian, Eastern Asian, Korean, Japanese, African, South American, American, Afro-American, Eastern European, European
Skin color: pale, fair, light, olive, medium, dark, dark brown
Pubic hair: hairy, abundant, well-trimmed, badly trimmed, shaved, well-shaved
Fluids (fluid dripping on her bodypart): milk, semen, cum, saliva, drool (I don't know AT ALL why it works but hey - it's the mistery of the model)
And a lot that I let you discover but here are the things I'm most focused on. I'm looking to do a similar jobs for mens, then combine the Loras, then train a checkpoint.
I hope you enjoy the model and if you do, please leave a like to give it visibility.
I have already a better version but as it's not much better I'm expanding the dataset before republishing.
Description
Better details, lower convergence so don't hesitate to push the lora a bit if it resists.
FAQ
Comments (10)
This new one seems very promising. Funny that given the attention, this type of LORA is usually the most lacking, so it's nice to see one that actually seems to get the job done- and with good poses to boot.
works great with flux fill
Some of those F2K previews look utterly real. 2026 is the year we can't tell the difference
Yeah crazy times, flexibility and edit abilities are becoming crazy too
@Holiwayz I haven't tried yet, I need a workflow. Is your included in the pictures? Thank you and good job, the LoRA looks amazing 🙌
@GlowingGuardianGirl Klein-9B (not base), 8 steps, cfg distilled 1 and you're good to go. from 1.5MP to 3MP~. Previews looks good when zoomed in, idk why the thumbnails looks so shitty
@Holiwayz [[Edit, I found the Huggingface link here.]]
Hey 😩. I set a workflow up, but I can't find a text encoder for 9B. Do you have any link? Thank you!
@GlowingGuardianGirl https://docs.comfy.org/tutorials/flux/flux-2-klein there are download links here
@Holiwayz Oh wow there's a lot of ressources for most models. Thanks for the link! 🙌
A little more fine tuning and Klein could easily become one of the greatest models to have come out (for the consumer level at least). The prompt coherence is excellent even for some complex prompts, and the near 1-to-1 ratio to actual photography without the need of using a detailing lora is just mind blowing. And this is coming from someone who started when SD 1.5 was the goat model.



















