Manrotica
[!NOTE] LoRA expertly trained by Divergent AI | Adult / NSFW
Full-frontal male erotica in editorial photographic grammar — solo nude spreads, honest erect and flaccid penis, and M/M softcore when you want it. Polaroid warmth and muscle-definition light without plastic skin.
✍️ A Note From the Creator
"Most male-focused checkpoints flatten skin into plastic and blur genitals into nothing — they dodge the photographic grammar that makes homoerotic work feel observed rather than generated. Manrotica was trained on 280 curated explicit photographs: solo spread-leg portraits with erect and flaccid penis visible, groin close-ups, shower and bed full nudity, open-mouthed kissing pairs, kneeling oral sex, ejaculate on skin. Write like a shoot director naming anatomy and acts, not a tag list, and this model will meet you there."
— Divergent AI
🎨 The Aesthetic
This is a contact sheet pulled from a photographer's private erotic archive — not a lighting demo. Silver monochrome torsos under hard sidelight, legs spread on a wooden bench with circumcised flaccid penis and scrotum between the thighs; colour frames where a pink paneled wall or blue pool lane saturates behind a lean nude with trimmed pubic hair. Sweat and shower water bead on defined abs and run down inner thigh toward the groin. Solo subjects hold the frame first — seated spreads, standing full-frontal, rear-view buttocks on beach recline, post-climax glaze on erect penis before a motel mirror. When two men share the frame, tongues meet in open-mouthed kiss, hands grip thigh near groin, knees drop for oral. Polaroid-style borders when you ask. Backgrounds blur so veins, pubic curl, and sand on bare skin carry the shot.
🔥 What's New in V2026.1?
👑 Solo Nude & Groin Fidelity: Spread-leg seated poses, standing full-frontal, and low-angle groin close-ups — erect and flaccid penis, scrotum, pubic hair rendered with photographic conviction
🍆 Circumcised & Uncircumcised Penis: Natural proportion and state across both — the model learned genitals as concepts, not afterthoughts
👨❤️👨 M/M Softcore Range: Open-mouthed kissing, kneeling oral sex, mutual groin touch, ejaculate on abdomen and glans — duo when you prompt it, solo by default
🎨 Polaroid Erotic vs Colour Bedroom/Shower/Pool: Black-and-white instant-film nude studies separate cleanly from saturated colour erotic interiors and wet skin
🧠 280 Explicit Caption-Aligned Photographs: Training on scene briefs that name penis, oral sex, and spread legs — not fantasy tag dumps
⚡ Key Features
Genital & Pubic Rendering: Erect and flaccid penis, scrotum, pubic hair — circumcised and uncircumcised — with vein definition and skin pore detail on the groin
Solo Spread-Leg & Groin Framing: Seated legs apart, reclining nude, low-angle close-up on penis and thighs — the dominant output mode
M/M Softcore Duo Scenes: Kissing with tongue, kneeling oral sex, partner hand on thigh near groin, semen on palm or glans when directed
Wet Skin & Steam: Shower cascade on hairy chest and erect penis, pool water on spread legs, sweat sheen on abs under sidelight
Triggerless Integration: No activation word — name the nude pose, genital state, and act in plain language; works across compatible photorealistic checkpoints
🚀 Ideal Use Cases
Solo Nude Portraits & Groin Close-Ups: Spread-leg seated poses, standing full-frontal, flaccid or erect penis visible, pubic hair and scrotum in frame
Shower, Bed & Mirror Erotic Studies: Full nudity with water on skin, white bedding, cracked motel mirror — solo self-styled male erotica
M/M Softcore Duo Scenes: Two nude men kissing, oral sex in steam, hands on groin, post-climax ejaculate detail
Polaroid B&W Nude Series & Intimate Video: Instant-film bordered spreads, monochrome muscle and genital studies, short-form reels with steam and body movement
⚙️ Quick Settings for Best Results
LoRA Strength:
1.0— do not go below 1.0. Manrotica encodes learned NSFW concepts — penis anatomy, erect and flaccid states, pubic hair, oral sex, ejaculate — that base checkpoints suppress or blur at partial strength. This is a concept realism adapter, not a light stylistic wash; lowering strength trades away the anatomical and behavioral vocabulary the dataset taught before it improves skin texture.Sampler Steps:
25 – 40for maximum skin and genital micro-detail on close-upsResolution: Flexible; tested across standard portrait and landscape dimensions up to ~1.3 MP
Base Model: Highly responsive on
{{base_model}}and premium photorealistic checkpointsContent Rating: Adult / NSFW — full nudity, explicit genitals, and softcore male-male sexual acts
❤️ Support The Project
👍 Like and Rate the model on the hosting platform
🖼 Share Your Creations! Post your best nudes and duo scenes in the gallery — seeing how you use it is the best motivation
💸 Tip the Creator to support ongoing training costs
🗣 Leave Feedback in the discussion section — your reviews steer future versions
Created by Divergent AI
Prompting Guide: Manrotica
Manrotica is a triggerless Adult/NSFW homoerotic realism engine. It was trained primarily on solo full nudity — spread-leg portraits, groin close-ups, erect and flaccid penis, scrotum, pubic hair — with M/M softcore (open-mouthed kissing, oral sex, mutual groin touch, ejaculate) when you direct it. Scene description drives output, not activation keywords. Write with the same direct anatomical vocabulary as the training captions (penis, scrotum, pubic hair, erect, flaccid, oral sex, ejaculate when the scene calls for it). Do not euphemize or imply what you can name explicitly.
🎯 What This Model Is For
Solo nude portraits — spread-leg seated poses, standing full-frontal, reclining on bed with penis visible
Groin close-ups — low angle on erect or flaccid penis, scrotum, pubic hair, inner thigh
Shower, bed & mirror erotic studies — full nudity, water on skin, steam, post-climax detail
M/M softcore duo scenes — kissing with tongue, kneeling oral sex, hands on thigh near groin, semen on glans or palm
Polaroid B&W nude series — instant-film bordered spreads and monochrome genital studies
🧠 Prompting Philosophy: Write Like a Novelist
To get the absolute best results, do not use simple keyword-stuffed tags. Paint a detailed picture with your words using these five core pillars:
The Subject: Specify build (lean, muscular, hairy), grooming (beard, pubic hair, tattoos), expression (neutral, half-lidded, open-mouthed), and genital state by name — circumcised or uncircumcised, erect or flaccid penis, scrotum visible, pubic hair density. Never leave the groin implied.
The Environment: Name materials — wooden bench, tiled shower, pink paneled wall, sandy beach pebbles, white bedding, cracked motel mirror — and whether the background is blurred, dark, or sunlit.
The Camera: Control angle for erotic framing — low groin close-up on penis and thighs, rear three-quarter on buttocks, seated eye-level with legs spread apart, overhead bed shot on two nude men. Focal isolation on genitals or kiss; Polaroid-style white border when wanted.
The Atmosphere: Direct light on bare skin — rim-lit doorway on erect penis, shower water cascade on hairy chest, golden rooftop haze on spread legs, dim lamp orange glow on inner thigh and scrotum — and tonal treatment (monochrome silver nude vs. saturated accent colour).
The Style: Anchor in photographic medium: homoerotic editorial photograph, black and white instant film nude, underwater study with erect penis visible through water, intimate close-up with skin pores and vein topography on the groin.
⚡ Putting It All Together: The Evolution
⛔ Simple/Okay Prompt (AI Slop / Keyword-Stuffed):
muscular man, nude, handsome, bedroom, photorealistic, 8k, detailed skin, perfect body✅ Detailed/Excellent Prompt (using Specificity Pillars):
Black and white photograph of a muscular, bearded man sitting on a wooden bench with legs spread apart, nude, wearing a chunky chain necklace. He has floral and skull tattoos across his chest and thighs. His circumcised, flaccid penis rests between his legs. He holds a serious expression, short dark hair catching a hard sidelight from a balcony railing behind him. Fine film grain and a Polaroid-style white border frame the urban exterior beyond.🔑 Trigger Words & Aesthetic Anchors
Trigger status: Triggerless — no activation keyword required. Do not prepend a LoRA name.
This vocabulary is how you direct nude bodies, genitals, and sexual acts — use it plainly in every prompt that should read as erotica, not shirtless editorial.
Subject & Form: muscular build, lean physique, defined abs, nude male, circumcised, uncircumcised, erect penis, flaccid penis, scrotum, buttocks, pubic hair, dorsal musculature, tattooed chest, short dark hair, beard, two men
Texture & Material: visible veins on penis shaft, sweat sheen, water droplets on skin and groin, sand grains on buttocks, skin pores, glistening wet pubic hair, ejaculate on glans, white tile grout
Lighting & Composition: soft side-lighting on spread legs, backlit doorway on nude torso, blurred neutral background, low angle groin close-up, legs spread apart, arms behind head, Polaroid style border, black and white photograph
Sexual Behavior: oral sex, open-mouthed kiss, tongue, hand on thigh near groin, kneeling, ejaculate, semen on abdomen, legs spread exposing penis
⚙️ Optimal Settings
LoRA Strength:
1.0— do not go below 1.0. NSFW concepts — penis, scrotum, erect and flaccid states, pubic hair, oral sex, ejaculate — are the core learned behaviors. Diluting strength causes base-model censorship, blurred genitals, or anatomical collapse. If close-ups over-sharpen, adjust steps, sampler, or base checkpoint — not strength.Sampler Steps:
25 – 40— higher steps preserve vein detail on penis shaft and shadow gradation on spread inner thigh.Negative Prompting: Use light negatives only (
blurry, deformed hands, extra limbs). Heavy negative stacks can flatten organic skin texture and suppress explicit anatomy this LoRA is meant to render.
🎬 Directing the Scene (Visual Vocabulary)
For Gaze & Presence: "looking directly at the camera with a neutral expression", "eyes closed, head tilted back", "half-lidded gaze over shoulder", "open-mouthed exhale, relaxed jaw", "tongues meeting in open-mouthed kiss"
For Physical Posture: "legs spread apart, hands resting on thighs, flaccid penis visible between legs", "one arm raised, hand behind head, erect penis jutting forward", "kneeling viewed from behind, feet crossed at ankles, buttocks and scrotum visible", "reclining with arms above head on pillow, pubic hair and penis exposed"
For Form & Lighting: "sunlight highlighting defined pectorals, abs, and erect penis", "rim light tracing the dorsal spine and buttocks cleft", "water pouring down tiled shower wall, droplets on hairy chest and groin", "dim warm lamp casting orange glow on inner thigh and scrotum"
For Sexual Behavior (Solo): "legs spread exposing erect uncircumcised penis and dense pubic hair", "hand resting on thigh near groin", "ejaculate on glans and lower abdomen", "flaccid circumcised penis resting on thigh after climax"
For Sexual Behavior (Duo): "kneeling oral sex on partner's erect penis", "open-mouthed kiss with hand on partner's thigh near groin", "semen on partner's palm", "two nude men on bed, one gripping the other's hip near groin"
💡 Pro-Tips for Best Results
Solo first: Lead with single-subject nude spreads and groin close-ups before duo scenes — roughly half the training set is solo full nudity. Name erect or flaccid penis and pubic hair in every explicit prompt.
Strength at 1.0 always: Run at full strength when prompting penis, oral sex, ejaculate, or spread-leg poses — partial strength is how base models win and genitals disappear.
Anatomy-to-environment ratio: Alternate close-up groin studies with full-environment duo scenes to avoid repetitive framing; the model generalizes well across both scales.
Diversity representation: Vary skin tone descriptors (light tan, medium brown, dark skin), hair texture (curly, short brown, shoulder-length blonde), and body hair density for range beyond the Caucasian-leaning training skew.
Monochrome vs. colour: Explicitly request "black and white photograph" or name a saturated accent environment (pink wall, blue pool lane) — the model bifurcates cleanly between tonal modes.
Prohibited AI Slop Adjectives: Avoid "stunning", "breathtaking", "ultra-detailed", "perfect body" — replace with concrete erotic texture: visible veins on penis shaft, Polaroid border, legs spread apart, sweat sheen on pectorals, ejaculate on glans.
Description
Technical Specifications
Base Model: Krea 2
Version: 2016.1
Recommended Weight: 1.0 (NO LESS)
Resolution: 512 (Trained) 1024 (Output)
CFG Scale: 2.5-4.5 (Base) 1.0 (Turbo)
Steps: 35-50 (Base) 8-9 (Turbo)
NSFW KEYWORDS:
Anatomy & nudity
nude maleerect penis/flaccid peniscircumcised/uncircumcisedscrotum/testiclespubic hair/dense pubic hair/trimmed pubic hairbuttocks/buttocks cleftgroin/inner thighglans/penis shaftvisible veins/visible veins on penis shaftdefined abs/dorsal musculature/hairy chest
Body & build
muscular build/lean physiquebeard/short dark hairtattooed chest/tattoos on thighshairy/smooth skin
Poses (solo-first)
legs spread apart/legs spread exposing penisseated nude/standing full-frontalreclining on bed/arms above head on pillowkneeling viewed from behindfeet crossed at ankleshand resting on thigh near groinarms behind headone arm raised, hand behind head
Sexual behavior (M/M softcore)
oral sex/kneeling oral sexopen-mouthed kiss/tongue/tongues meetingejaculate/semen on abdomen/semen on glans/semen on palmpost-climax/after climaxhand on thigh near groin/gripping hip near grointwo men/two nude men
FAQ
Comments (12)
请问对亚洲男性,还有古风,是否有同样的优秀效果
嗨,谢谢你对我的 LoRA 感兴趣。不幸的是,我觉得这个模型在处理亚洲面孔时效果可能不会特别好。我会在以后努力扩大族裔数据集。不过你可以尝试一下,其实说不定还是有用的,只是可能只有一半的成功率。不好意思啦,等我有时间会优化这个问题的!
Great! This will be so usefull. Thanks for sharing!
Only a pleasure, please make sure to post your images if you can 😜
Hey! Great checkpoint.
Can you tell me more about how you train flaccid and erect together? In my own experience it always shows one type after training, whatever I tried so far.
But maybe you captioned them in a certain way that it does work?
Tagging with flaccid penis vs erect penis does not work here at all when training.
Thanks in advance if you could give some guidance!
Hey so I can't tell you exactly what the magic juice is. Mainly because I'm not sure what makes it work. But I think it's a combination of the below.
I hope you're ready for detail lol. LoRA training/ML is a bit of an obsession to me.
1. Large dataset
- All models I train are usually style and trained on datasets that would be considered huge to some. 200-1000+. My MASC model is something close to 2k. This model is around 500 with only half being explicit.
- The mix I like is like:
65-80% the concept or style you're trying to train
10-15% adjacent concepts you think your concept will utilize often (like leather bars if you're doing a leather LoRA for example, or images of leather detailing on fashion or some editorial bondage fashionwear. Those are all going to add to your LoRAs capabilities)
10-20% concepts in the same style. So for example this is a realistic style right, so I trained it on images of skin textures and realistic photos of general human beings. This ensures the model doesn't overfit to a specific "type" and allows you to generate stuff outside of your dataset. This is great for example if you have mostly Caucasian men in a specific type of imagery but lack diverse subjects. You could then fix this by training on your caucasian images that are NSFW AND training on SFW diverse raced people. Your model will be a lot better, not perfect, but better at generating other ethnicities in the NSFW way.
... I should probably come up with an acronym for it.
- This doesn't mean it's impossible with smaller datasets, I just push for larger ones because I'm a stickler for quality
2. HIGHLY detailed captions
- My models don't require activation words because I intentionally prompt very verbose (using AI). I find this make the model take way better action overall on diverse prompts.
-- Simple math in my head, the more tokens you have in your captions, the more likely a caption someone writes is going to hit something that's appeared in the dataset.
-- Start with a short concise sentence or your trigger word, so that those still work, then follow with highly detailed caption
-- This could be why it innately learns to separate different concepts like uncut vs cut.
- I often have captions that are paragraphs long, sometimes well above 300 tokens. I love it. I'd never prompt like that without AI. But I find it does wonders for LoRA training.
3. Multi-stage training
- I don't always do this as it's not necessary, for example MASC and all VV series on Krea 2 are done with a single training run.
-- For this one though I tried something special which I might update to my other workflows as well.
--- I pushed my training into 3 parts.
Part 1:
I focused only on high noise, did 225px resolution (yes that low, going higher is a waste we're not capturing detail here), batch size 2, though that I don't think is that necessary. I'm just extra.
This forces the model to learn the context of the concept. This is especially important when pushing for NSFW content because you're really trying to push the model out of it's safety guidance, yes even Krea 2 has those.
I think if anything this might have been the deciding factor because it pushed the model to learn the difference between the two.
Part 2:
I then run the training from it's last checkpoint but now I change it usually to whatever the models default settings are. I take the batch down to 1 and I take the resolution up to 512.
Part 3:
Just because I focused on realism and I think it's important for this LoRA specifically because NSFW anatomy has VERY little MALE data in the image models. I further finetune on LOW noise and HALF learning rate for a few (like +/- 2500) thousand steps.
Amount of steps:
I think it's also important to note that the amount of steps is very tied to the size of the dataset. If the dataset is smaller it would obviously require less steps and overtrain otherwise. So be cautious.
For this lora it was like this:
Part 1: 3000 steps LR 0.0001 (1e-04), high noise weighted
Part 2: 2000 steps LR 0.0001 (1e-04), I think it was linear balanced noise
Part 3: 2500 steps LR 0.00005 (5e-05), low noise weighted
Thank you for coming to my TED Talk.
Hey sorry to come back after such a long comment, but I was just thinking you can actually even see the limitation of the diversity of my dataset in this LoRA.
I'm actually planning to far extend it. You'll notice this model struggles to generate hardcore concepts (which I'm quite ok with), other ethnicities might be a miss, more than two people not amazing. It also seems to generate to very similar people but that could just be the turbo pushing the diversity down.
Let me know if you notice any other failure cases.
I think I'm going to make a more realistic focused model too sometime. This one is VERY editorial.
@thedivergentai Thanks for the long detailled comment, really appreciated!! I turned on Differential Guidance at 3 the last run and it worked already better, also seems to follow my prompts way better without safety bypass lora's needed. The same dataset, description and settings made a big difference with this. Somehow though, the downside, less realisme with DG on.
But will definitely try some of your settings here!
do you know why there are artefacts especially when there is blank background?
Could you dm me an example so I can better understand? You could try turn it down to 0.9 which might fix the artifacts but might also lead to NSFW misses. I'll see if I can sort that out for a next release.
@thedivergentai i.g. this prompt sometime gives artifacts on skin and the background looks very low quality (it's like iso1600) "three men standing in a studio. They are posing isolated on a clean, solid blue background. They are completely nude except for sneakers, with their penises fully exposed. They have deep tanned skin and average bodies. They are posing in fashion confident model style.
The man on the left has a Top knot haircut.
The man in the center has Edgy Hard Part Fade hair.
The man on the right has a Undercut with long curly fringe hairstyle.
They are all looking toward the viewer with confident expressions. The lighting is bright and even studio lighting."
Oh sorry, how silly of me not to ask. What settings are you using for generation? I'll try have a look with your prompt when I get a chance. I have an idea what it may be. And it's something I'm already planning to augment in future releases. So we'll see. Thanks for the feedback though! Will come back when I have updates.


















