Downloads · 30 days
1.5K
40% of all-time downloads
perpetual3x/Qwen-Image-Korean-Female-Anatomy-Realism-NSFW-LoRA-v1
Qwen-Image-Korean-Female-Anatomy-Realism-NSFW-LoRA-v1 is a text-to-image model from perpetual3x. Use it when you need an image from a text prompt. It is set up for diffusers. The card lists the license as other.
<div style="display:flex; flex-wrap:nowrap; gap:6px; align-items:center; overflow-x:auto;"<a href="" style="display:inline-flex; align-items:center; height:22px; vertical-align:middle;"<img alt="Visitors" src="https:/…
Downloads · 30 days
1.5K
40% of all-time downloads
All-time downloads
3.7K
Public
Repo size
609 MB
Likes
5
Public
Click a slice to open those files.
.safetensors590 MB · 97%
From the Hugging Face model README

A note about language: This release concerns Korea, but it does not include a Korean-language edition—not because Korean matters less, but because I do not speak it. English is my native language; I also speak French and Spanish, and I have studied Japanese and Chinese at a beginner level. Korean, however, remains outside my present competence. Producing a confident machine translation would be easy. Publishing it as though it were my own voice would be less respectful than admitting the boundary. If a Korean edition appears later, it should be prepared or carefully reviewed by someone who actually knows where every particle belongs.
🇬🇧 English - displayed below
Note linguistique : cette publication concerne la Corée, mais ne propose pas d'édition coréenne — non parce que le coréen aurait moins d'importance, mais parce que je ne le parle pas. L'anglais est ma langue maternelle ; je parle aussi français et espagnol, et j'ai étudié le japonais et le chinois à un niveau débutant. Le coréen reste toutefois hors de mes compétences actuelles. Produire une traduction automatique pleine d'assurance serait facile. La publier comme si elle exprimait réellement ma voix serait moins respectueux que de reconnaître cette limite. Si une édition coréenne paraît un jour, elle devra être préparée ou soigneusement relue par quelqu'un qui sait vraiment où placer chaque particule.
Avertissement — contenu pour adultes : ce dépôt contient des images explicites générées par IA et s'adresse exclusivement aux adultes. Toutes les personnes décrites dans les exemples et les prompts de référence sont adultes.

Il s'agit de la dernière publication de ma série consacrée à l'anatomie et au réalisme humain. L'Europe occidentale et l'Amérique ont ouvert le chapitre visuel le plus large ; le Japon et la Chine ont suivi avec leurs études régionales de l'anatomie féminine ; le projet d'anatomie masculine a fourni la moitié de l'humanité que les modèles d'image réduisent trop souvent soit au marbre poli, soit à une plomberie improvisée. La Corée verrouille maintenant la cinquième et dernière pièce. Elle n'arrive pas en dernier parce qu'elle compte moins. Elle arrive en dernier parce que tout puzzle exige une pièce qui rend soudain l'image complète - et qu'après avoir examiné autant d'anatomie générée, l'auteur comme le GPU Blackwell ont gagné le droit de cesser de prétendre que la topologie est un loisir.
La série s'arrête ici de façon délibérée. Chaque modèle pourra encore s'améliorer, mais l'ensemble prévu est complet : réalisme occidental, anatomie japonaise, chinoise, coréenne et masculine occupent désormais la même image achevée. Pendant quelque temps, je compte reposer mes yeux, sortir et regarder des paysages où personne n'a besoin d'un numéro de seed, d'un poids d'adaptateur ou d'une discussion détaillée sur la continuité des tissus.
La culture visuelle coréenne passe naturellement des cours de hanok aux ateliers de céladon, des marais côtiers aux temples de montagne, de la pierre assombrie par la pluie aux intérieurs contemporains et aux lumières urbaines denses. Un modèle capable de restituer tout cela ne devrait pas atteindre le corps humain pour remplacer soudain l'observation par un espace réservé lisse et vaguement sûr de lui.
Ce LoRA se concentre sur l'anatomie adulte réaliste d'Asie de l'Est, avec un accent d'entraînement particulièrement marqué sur les femmes coréennes et le contexte photographique coréen. Cela décrit le jeu de données et sa distribution visuelle ; cela ne prétend pas que des millions de personnes partagent un seul visage, un seul corps ou une seule configuration d'usine.
Les systèmes de beauté savent très bien polir les surfaces. L'anatomie pose une question moins glamour : tout reste-t-il relié, proportionné et physiquement plausible après les négociations entre lumière, pose, gravité et angle de caméra ? Texture de peau, plis, compression, pilosité et asymétrie ordinaire ne sont pas des erreurs de rendu. Ce sont les preuves que le corps a été autorisé à rester un corps.
Ce modèle vise donc une anatomie externe crédible plutôt qu'une perfection cosmétique. Il cherche à préserver la continuité visuelle entre la vulve, la région clitoridienne, le périnée, l'anus, le bassin et les tissus environnants, aussi bien dans les études anatomiques directes que dans les compositions environnementales plus larges.
Il s'agit d'un LoRA d'anatomie et de réalisme. Il peut fonctionner seul lorsque le prompt définit le sujet adulte, la scène et la composition, ou comme spécialiste aux côtés d'un autre LoRA NSFW. En combinant plusieurs adaptateurs, laissez le LoRA principal établir l'identité ou le style et n'utilisez celui-ci qu'à l'intensité nécessaire pour restaurer la cohérence anatomique.
| Élément | Valeur |
|---|---|
| Modèle de base | Qwen Image 2512 |
| Images d'entraînement | 5 822 |
| GPU | NVIDIA RTX PRO 6000 Blackwell / 96 GB |
| Rank | 32 |
| Taille de lot | 8 |
| Étapes d'entraînement | 10 000 |
| Déclencheur | Aucun — langage naturel uniquement |
0.8-1.2, puis adaptez-le à la pose, à la distance de caméra, à la seed et aux autres LoRA actifs0.8, préservez d'abord l'identité et le langage visuel du LoRA principal, puis augmentez ce modèle seulement jusqu'à stabilisation de la correction anatomiqueLe jeu d'entraînement est nettement plus large que celui des précédentes publications régionales, mais 5 822 images n'abrogent toujours ni la géométrie, ni la gravité, ni les mauvaises seeds.
Tous les exemples représentent des femmes clairement adultes. Le prompt de couverture n'est pas fourni. Les huit cas sont reproduits ci-dessous ; leurs prompts exacts, conservés en anglais pour une utilisation directe, figurent dans la version anglaise affichée plus bas.
| Exemple généré |
|---|
![]() |
Prompt
a clearly adult Korean woman age 49 with a slender mature build, a white hemp hanbok robe hanging open and secured behind the hips so no fabric crosses the pelvis, leaning forward with both hands on a low soban table while both feet remain grounded in a pottery workshop in Icheon, surrounded by celadon vessels, clay tools and kiln light. a ground-level rear view looks gently upward beneath the pelvis without hiding the surrounding scene; the anal opening is genuinely readable rather than implied by the gluteal cleft, and the perineum continues coherently toward the vulva. summer sunlight broken by leaves and moving clouds, natural editorial photography with precise anatomy and gentle depth of field; a traditional embroidered cushion and bojagi wrapping cloth add controlled color. Accurate clearly adult proportions, realistic hands and feet, no additional people, no text, no logo, no watermark.
| Exemple généré |
|---|
![]() |
Prompt
a clearly adult Korean woman age 43 with a compact muscular build, a crimson silk chima untied and pooled around her feet, its matching jeogori worn open, lying on her left side with the lower leg extended and the upper leg supported on a cushion near the reeds of Suncheon Bay, with distant wetlands, birds and a broad luminous sky. a wide lateral composition shows the full body line and cultural setting without obstruction; the vulva, clitoral hood and perineum remain coherent in profile with realistic labial compression and pubic attachment. gentle overcast daylight preserving realistic skin texture, contemporary fine-art realism with authentic material textures and soft asymmetry; a small vase of plum blossoms echoes the season without distracting from the figure. Accurate clearly adult proportions, realistic hands and feet, no additional people, no text, no logo, no watermark.
| Exemple généré |
|---|
![]() |
Prompt
This intimate documentary-style image records an adult nude Korean woman with realistic proportions and natural skin texture resting on both hands on a long wooden table while turning toward an unseen sound inside a tiled winter garden with citrus trees and a long iron bench. Her softly tousled hair reinforces the unpolished mood. The viewpoint is frontal but slightly off-center, leaving room for reflected light and negative space. North-window light falls softly across the body and leaves the room in gentle shadow. The visual language is cinematic but intimate, with gentle lens softness and believable environmental detail. The image feels intimate without becoming glossy or theatrical. Avoid plastic skin, exaggerated symmetry or fashion retouching; retain natural variation and small imperfections.
| Exemple généré |
|---|
![]() |
Prompt
This poetic environmental portrait features an adult nude Korean woman with a natural, unretouched appearance resting on both hands on a long wooden table while turning toward an unseen sound inside a faded hotel room with patterned wallpaper and a tall mirror. Her hair is gathered away from her face, leaving the gesture clearly readable. The camera holds a natural eye-level perspective with no exaggerated foreshortening. Backlight through linen curtains outlines the figure while reflected fill retains detail. Its palette is desaturated and painterly, with gentle contrast and no commercial retouching. The atmosphere suggests solitude and an unforced connection with the surrounding place. Preserve full-body continuity, natural hands and feet, plausible weight distribution and understated expression.
| Exemple généré |
|---|
![]() |
Prompt
A restrained editorial figure study depicts an adult nude Korean woman integrated quietly into the surrounding environment resting on a low stone wall with both feet touching the ground and her torso turned toward the horizon at the edge of an abandoned orchard overlooking a wide agricultural plain. A few damp strands of hair rest naturally against her face and shoulders. A broad side-on composition emphasizes rhythm, balance and environmental texture. Open shade beneath the trees provides even illumination while the distance remains luminous. The print-like finish preserves natural asymmetry, folds and minor real-world imperfections. The emotional tone is reflective and slightly melancholic. Maintain one coherent adult body, natural joints, believable balance and realistic proportions.
| Exemple généré |
|---|
![]() |
Prompt
The image presents an adult nude Korean woman whose presence feels candid and unperformed lying across a smooth boulder warmed by the remaining daylight on an orchard path covered with fallen blossoms after a spring shower. Her softly tousled hair reinforces the unpolished mood. A distant medium-format composition lets weather and place occupy most of the frame. Soft overcast daylight wraps the figure without hard highlights. The result resembles restrained 35 mm color-negative film, with fine grain, muted saturation and honest texture. The gesture reads as a private moment observed with care and distance. Keep the gesture physically achievable and the relationship between body, gravity and support convincing.
| Exemple généré |
|---|
![]() |
Prompt
This poetic environmental portrait features an adult nude Korean woman treated as a classical figure study reclining on a long bench beneath tall windows and damaged plaster inside an abandoned winter pavilion with cracked tiles and fogged glass. Her loose hair follows the turn of her head. The camera looks through foreground leaves or fabric, adding a restrained sense of depth. Pale winter light enters through damaged glass and settles quietly across the architecture. The print-like finish preserves natural asymmetry, folds and minor real-world imperfections. The environment and figure share equal visual importance, producing a calm sense of scale. Maintain one coherent adult body, natural joints, believable balance and realistic proportions.
| Exemple généré |
|---|
![]() |
Prompt
A museum-minded portrait observes an adult nude Korean woman integrated quietly into the surrounding environment reclining on a long bench beneath tall windows and damaged plaster beneath a concrete colonnade after rain, with reflections across the floor. Her hair is tied back in a simple, practical knot. A waist-level side view follows the natural line of the gesture. Late-day light catches the upper walls while reflected fill remains soft below. The visual language is cinematic but intimate, with gentle lens softness and believable environmental detail. The composition balances bodily presence with landscape-driven stillness. Maintain one coherent adult body, natural joints, believable balance and realistic proportions.
Le fichier téléchargeable krvulva_traditional_korea_1000.txt contient 1 000 prompts couvrant vêtements traditionnels, intérieurs, paysages, ateliers, temples, littoraux, lumières saisonnières et tests de caméra centrés sur l'anatomie. Ce sont des matériaux de test, pas la promesse que chaque seed se comportera en citoyenne exemplaire.
Cette publication coréenne place la cinquième et dernière pièce dans la série consacrée à l'anatomie et au réalisme humain : occidental, japonais, chinois, coréen et masculin. J'ai désormais passé assez de temps à comparer seeds, géométrie du bassin, lumière environnementale et opinions privées de la gravité pour mériter une courte période de chômage visuel volontaire. Les modèles peuvent continuer à générer ; moi, je sors. Si vous avez une idée réellement nouvelle, un rapport d'échec utile ou une direction digne d'être explorée après cette pause, vous pouvez me contacter via le dépôt. Les bonnes idées restent intéressantes même lorsque l'auteur refuse temporairement d'inspecter quoi que ce soit à 800 % de zoom.
Les fichiers téléchargeables reference.py et reference_low_vram.py chargent Qwen-Image-2512 en BF16, fusionnent l'adaptateur Lightning communautaire et appliquent ce LoRA coréen sans déclencheur. La version standard recommande un GPU CUDA à grande mémoire.
pip install -U diffusers transformers accelerate peft safetensors
<details>
<summary><strong>▶ Afficher le code Python standard</strong></summary>
"""Qwen Image 2512 + Korean Female Anatomy Realism LoRA v1."""
import math
import torch
from diffusers import DiffusionPipeline, FlowMatchEulerDiscreteScheduler
BASE_MODEL_ID = "Qwen/Qwen-Image-2512"
LIGHTNING_MODEL_ID = "lightx2v/Qwen-Image-2512-Lightning"
LIGHTNING_WEIGHT_NAME = (
"Qwen-Image-2512-Lightning-4steps-V1.0-bf16.safetensors"
)
LORA_MODEL_ID = (
"perpetual3x/Qwen-Image-Korean-Female-Anatomy-Realism-NSFW-LoRA-v1"
)
LORA_WEIGHT_NAME = "krvulva.safetensors"
DEVICE = "cuda"
DTYPE = torch.bfloat16
LORA_SCALE = 0.9
scheduler_config = {
"base_image_seq_len": 256,
"base_shift": math.log(3),
"invert_sigmas": False,
"max_image_seq_len": 8192,
"max_shift": math.log(3),
"num_train_timesteps": 1000,
"shift": 1.0,
"shift_terminal": None,
"stochastic_sampling": False,
"time_shift_type": "exponential",
"use_beta_sigmas": False,
"use_dynamic_shifting": True,
"use_exponential_sigmas": False,
"use_karras_sigmas": False,
}
scheduler = FlowMatchEulerDiscreteScheduler.from_config(scheduler_config)
pipe = DiffusionPipeline.from_pretrained(
BASE_MODEL_ID,
scheduler=scheduler,
torch_dtype=DTYPE,
).to(DEVICE)
pipe.load_lora_weights(
LIGHTNING_MODEL_ID,
weight_name=LIGHTNING_WEIGHT_NAME,
adapter_name="lightning",
)
pipe.fuse_lora(adapter_names=["lightning"], lora_scale=1.0)
pipe.unload_lora_weights()
pipe.load_lora_weights(
LORA_MODEL_ID,
weight_name=LORA_WEIGHT_NAME,
adapter_name="korean_anatomy_v1",
)
pipe.set_adapters(["korean_anatomy_v1"], adapter_weights=[LORA_SCALE])
prompt = (
"A realistic photograph of a clearly adult Korean woman, coherent "
"external female anatomy, natural pelvic structure, unretouched skin "
"texture, physically plausible pose, soft natural light"
)
negative_prompt = (
"child, minor, underage, youthful, low quality, blurry, distorted anatomy, "
"extra limbs, duplicated body parts, fused anatomy, watermark, logo, text"
)
generator = torch.Generator(device=DEVICE).manual_seed(42)
image = pipe(
prompt=prompt,
negative_prompt=negative_prompt,
width=928,
height=1328,
num_inference_steps=8,
true_cfg_scale=1.0,
generator=generator,
).images[0]
image.save("korean_female_anatomy_realism_v1.png")
</details>
<details>
<summary><strong>▶ Afficher le code Python à faible VRAM</strong></summary>
"""Low-VRAM Qwen Image 2512 + Korean Female Anatomy Realism LoRA v1."""
import gc
import math
import torch
from diffusers import DiffusionPipeline, FlowMatchEulerDiscreteScheduler
BASE_MODEL_ID = "Qwen/Qwen-Image-2512"
LIGHTNING_MODEL_ID = "lightx2v/Qwen-Image-2512-Lightning"
LIGHTNING_WEIGHT_NAME = (
"Qwen-Image-2512-Lightning-4steps-V1.0-bf16.safetensors"
)
LORA_MODEL_ID = (
"perpetual3x/Qwen-Image-Korean-Female-Anatomy-Realism-NSFW-LoRA-v1"
)
LORA_WEIGHT_NAME = "krvulva.safetensors"
DTYPE = torch.bfloat16
LORA_SCALE = 0.9
scheduler_config = {
"base_image_seq_len": 256,
"base_shift": math.log(3),
"invert_sigmas": False,
"max_image_seq_len": 8192,
"max_shift": math.log(3),
"num_train_timesteps": 1000,
"shift": 1.0,
"shift_terminal": None,
"stochastic_sampling": False,
"time_shift_type": "exponential",
"use_beta_sigmas": False,
"use_dynamic_shifting": True,
"use_exponential_sigmas": False,
"use_karras_sigmas": False,
}
gc.collect()
if torch.cuda.is_available():
torch.cuda.empty_cache()
scheduler = FlowMatchEulerDiscreteScheduler.from_config(scheduler_config)
pipe = DiffusionPipeline.from_pretrained(
BASE_MODEL_ID,
scheduler=scheduler,
torch_dtype=DTYPE,
low_cpu_mem_usage=True,
)
pipe.load_lora_weights(
LIGHTNING_MODEL_ID,
weight_name=LIGHTNING_WEIGHT_NAME,
adapter_name="lightning",
)
pipe.fuse_lora(adapter_names=["lightning"], lora_scale=1.0)
pipe.unload_lora_weights()
pipe.load_lora_weights(
LORA_MODEL_ID,
weight_name=LORA_WEIGHT_NAME,
adapter_name="korean_anatomy_v1",
)
pipe.set_adapters(["korean_anatomy_v1"], adapter_weights=[LORA_SCALE])
pipe.enable_sequential_cpu_offload()
pipe.vae.enable_tiling()
pipe.vae.enable_slicing()
prompt = (
"A realistic photograph of a clearly adult Korean woman, coherent "
"external female anatomy, natural pelvic structure, unretouched skin "
"texture, physically plausible pose, soft natural light"
)
negative_prompt = (
"child, minor, underage, youthful, low quality, blurry, distorted anatomy, "
"extra limbs, duplicated body parts, fused anatomy, watermark, logo, text"
)
generator = torch.Generator(device="cpu").manual_seed(42)
image = pipe(
prompt=prompt,
negative_prompt=negative_prompt,
width=768,
height=1024,
num_inference_steps=8,
true_cfg_scale=1.0,
generator=generator,
).images[0]
image.save("korean_female_anatomy_realism_v1_low_vram.png")
</details>
Ce modèle est exclusivement destiné à un travail créatif adulte, légal et consensuel. Ne l'utilisez pas pour sexualiser des mineurs ou des personnes d'apparence juvénile, créer du contenu non consenti ou exploiteur, représenter une personne réelle identifiable sans autorisation, harceler ou diffamer, produire du contenu illégal ou contourner les règles d'une plateforme.
Il incombe aux utilisateurs de s'assurer que leurs prompts, leurs sources et leurs contenus générés respectent les lois applicables ainsi que les règles des plateformes.
Ce LoRA peut être utilisé uniquement à des fins personnelles et non commerciales. Consultez le fichier LICENSE du dépôt pour connaître l'intégralité des conditions.
Pour toute demande de licence commerciale, tout retour technique utile ou toute idée de projet réellement nouvelle, veuillez contacter le propriétaire du dépôt.
</details> <details name="language-version"> <summary><strong>🌎 Español latinoamericano - mostrar la versión completa</strong></summary>Nota sobre el idioma: esta publicación trata de Corea, pero no incluye una edición en coreano; no porque el idioma importe menos, sino porque no lo hablo. El inglés es mi lengua materna; también hablo francés y español, y he estudiado japonés y chino a nivel principiante. El coreano, sin embargo, queda fuera de mis competencias actuales. Producir una traducción automática llena de seguridad sería fácil. Publicarla como si expresara mi propia voz sería menos respetuoso que reconocer el límite. Si aparece una edición coreana más adelante, debería prepararla o revisarla cuidadosamente alguien que realmente sepa dónde corresponde cada partícula.
Advertencia de contenido para adultos: este repositorio contiene imágenes explícitas generadas por IA y está destinado exclusivamente a adultos. Todas las personas descritas en los ejemplos y prompts de referencia son adultas.

Esta es la publicación final de mi serie de anatomía y realismo humano. Europa Occidental y Estados Unidos abrieron el capítulo visual más amplio; Japón y China siguieron con estudios regionales de anatomía femenina; y el proyecto de anatomía masculina aportó la mitad de la humanidad que los modelos de imagen reducen con demasiada frecuencia a mármol pulido o fontanería improvisada. Corea encaja ahora la quinta y última pieza. No quedó al final porque importe menos, sino porque todo rompecabezas necesita una pieza que haga que la imagen se lea de pronto como completa; después de examinar tanta anatomía generada, tanto el autor como la GPU Blackwell se han ganado el derecho a dejar de fingir que la topología es una actividad recreativa.
La serie termina aquí de forma deliberada. Cada modelo podrá seguir mejorando, pero el conjunto previsto está completo: realismo occidental y anatomía japonesa, china, coreana y masculina ocupan ahora la misma imagen terminada. Durante un tiempo pienso descansar la vista, salir y contemplar paisajes en los que nadie necesite un número de seed, un peso de adaptador ni una discusión detallada sobre continuidad de tejidos.
La cultura visual coreana se mueve con naturalidad entre patios hanok, talleres de celadón, humedales costeros, templos de montaña, piedra oscurecida por la lluvia, interiores modernos y luces urbanas densas. Un modelo capaz de reproducir todo eso no debería llegar al cuerpo humano y sustituir de pronto la observación por un marcador liso y vagamente seguro de sí mismo.
Este LoRA se centra en anatomía adulta realista de Asia Oriental, con especial énfasis de entrenamiento en mujeres coreanas y contexto fotográfico coreano. Eso describe el conjunto de datos y su distribución visual; no afirma que millones de personas compartan un rostro, un cuerpo o una configuración de fábrica.
Los sistemas de belleza son excelentes para pulir superficies. La anatomía hace una pregunta menos glamurosa: ¿todo sigue conectado, proporcionado y físicamente plausible después de que iluminación, pose, gravedad y ángulo de cámara hayan terminado de negociar? La textura de la piel, los pliegues, la compresión, el vello corporal y la asimetría cotidiana no son fallos de renderizado. Son pruebas de que al cuerpo se le permitió seguir siendo cuerpo.
Por ello, este modelo busca anatomía externa creíble en lugar de perfección cosmética. Está entrenado para conservar la continuidad visual entre vulva, región clitoriana, perineo, ano, pelvis y tejidos circundantes, tanto en estudios anatómicos directos como en composiciones ambientales más amplias.
Este es un LoRA de anatomía y realismo. Puede funcionar solo cuando el prompt aporta el sujeto adulto, la escena y la composición, o actuar como especialista junto a otro LoRA NSFW. Al combinar adaptadores, permita que el LoRA principal establezca la identidad o el estilo y use este modelo solo con la intensidad necesaria para recuperar la coherencia anatómica.
| Elemento | Valor |
|---|---|
| Modelo base | Qwen Image 2512 |
| Imágenes de entrenamiento | 5,822 |
| GPU | NVIDIA RTX PRO 6000 Blackwell / 96 GB |
| Rank | 32 |
| Tamaño de lote | 8 |
| Pasos de entrenamiento | 10,000 |
| Activación | Ninguna — solo lenguaje natural |
0.8-1.2 y ajústelo según pose, distancia de cámara, seed y demás LoRA activos0.8, conserve primero la identidad y el lenguaje visual del LoRA principal y aumente este modelo solo hasta estabilizar la corrección anatómicaEl conjunto de entrenamiento es mucho más amplio que el de las publicaciones regionales anteriores, pero 5,822 imágenes todavía no derogan la geometría, la gravedad ni las malas seeds.
Todos los ejemplos representan mujeres claramente adultas. El prompt de portada no se incluye. Los ocho casos se reproducen a continuación; sus prompts exactos, conservados en inglés para poder utilizarlos directamente, aparecen en la versión inglesa mostrada más abajo.
| Ejemplo generado |
|---|
![]() |
Prompt
a clearly adult Korean woman age 49 with a slender mature build, a white hemp hanbok robe hanging open and secured behind the hips so no fabric crosses the pelvis, leaning forward with both hands on a low soban table while both feet remain grounded in a pottery workshop in Icheon, surrounded by celadon vessels, clay tools and kiln light. a ground-level rear view looks gently upward beneath the pelvis without hiding the surrounding scene; the anal opening is genuinely readable rather than implied by the gluteal cleft, and the perineum continues coherently toward the vulva. summer sunlight broken by leaves and moving clouds, natural editorial photography with precise anatomy and gentle depth of field; a traditional embroidered cushion and bojagi wrapping cloth add controlled color. Accurate clearly adult proportions, realistic hands and feet, no additional people, no text, no logo, no watermark.
| Ejemplo generado |
|---|
![]() |
Prompt
a clearly adult Korean woman age 43 with a compact muscular build, a crimson silk chima untied and pooled around her feet, its matching jeogori worn open, lying on her left side with the lower leg extended and the upper leg supported on a cushion near the reeds of Suncheon Bay, with distant wetlands, birds and a broad luminous sky. a wide lateral composition shows the full body line and cultural setting without obstruction; the vulva, clitoral hood and perineum remain coherent in profile with realistic labial compression and pubic attachment. gentle overcast daylight preserving realistic skin texture, contemporary fine-art realism with authentic material textures and soft asymmetry; a small vase of plum blossoms echoes the season without distracting from the figure. Accurate clearly adult proportions, realistic hands and feet, no additional people, no text, no logo, no watermark.
| Ejemplo generado |
|---|
![]() |
Prompt
This intimate documentary-style image records an adult nude Korean woman with realistic proportions and natural skin texture resting on both hands on a long wooden table while turning toward an unseen sound inside a tiled winter garden with citrus trees and a long iron bench. Her softly tousled hair reinforces the unpolished mood. The viewpoint is frontal but slightly off-center, leaving room for reflected light and negative space. North-window light falls softly across the body and leaves the room in gentle shadow. The visual language is cinematic but intimate, with gentle lens softness and believable environmental detail. The image feels intimate without becoming glossy or theatrical. Avoid plastic skin, exaggerated symmetry or fashion retouching; retain natural variation and small imperfections.
| Ejemplo generado |
|---|
![]() |
Prompt
This poetic environmental portrait features an adult nude Korean woman with a natural, unretouched appearance resting on both hands on a long wooden table while turning toward an unseen sound inside a faded hotel room with patterned wallpaper and a tall mirror. Her hair is gathered away from her face, leaving the gesture clearly readable. The camera holds a natural eye-level perspective with no exaggerated foreshortening. Backlight through linen curtains outlines the figure while reflected fill retains detail. Its palette is desaturated and painterly, with gentle contrast and no commercial retouching. The atmosphere suggests solitude and an unforced connection with the surrounding place. Preserve full-body continuity, natural hands and feet, plausible weight distribution and understated expression.
| Ejemplo generado |
|---|
![]() |
Prompt
A restrained editorial figure study depicts an adult nude Korean woman integrated quietly into the surrounding environment resting on a low stone wall with both feet touching the ground and her torso turned toward the horizon at the edge of an abandoned orchard overlooking a wide agricultural plain. A few damp strands of hair rest naturally against her face and shoulders. A broad side-on composition emphasizes rhythm, balance and environmental texture. Open shade beneath the trees provides even illumination while the distance remains luminous. The print-like finish preserves natural asymmetry, folds and minor real-world imperfections. The emotional tone is reflective and slightly melancholic. Maintain one coherent adult body, natural joints, believable balance and realistic proportions.
| Ejemplo generado |
|---|
![]() |
Prompt
The image presents an adult nude Korean woman whose presence feels candid and unperformed lying across a smooth boulder warmed by the remaining daylight on an orchard path covered with fallen blossoms after a spring shower. Her softly tousled hair reinforces the unpolished mood. A distant medium-format composition lets weather and place occupy most of the frame. Soft overcast daylight wraps the figure without hard highlights. The result resembles restrained 35 mm color-negative film, with fine grain, muted saturation and honest texture. The gesture reads as a private moment observed with care and distance. Keep the gesture physically achievable and the relationship between body, gravity and support convincing.
| Ejemplo generado |
|---|
![]() |
Prompt
This poetic environmental portrait features an adult nude Korean woman treated as a classical figure study reclining on a long bench beneath tall windows and damaged plaster inside an abandoned winter pavilion with cracked tiles and fogged glass. Her loose hair follows the turn of her head. The camera looks through foreground leaves or fabric, adding a restrained sense of depth. Pale winter light enters through damaged glass and settles quietly across the architecture. The print-like finish preserves natural asymmetry, folds and minor real-world imperfections. The environment and figure share equal visual importance, producing a calm sense of scale. Maintain one coherent adult body, natural joints, believable balance and realistic proportions.
| Ejemplo generado |
|---|
![]() |
Prompt
A museum-minded portrait observes an adult nude Korean woman integrated quietly into the surrounding environment reclining on a long bench beneath tall windows and damaged plaster beneath a concrete colonnade after rain, with reflections across the floor. Her hair is tied back in a simple, practical knot. A waist-level side view follows the natural line of the gesture. Late-day light catches the upper walls while reflected fill remains soft below. The visual language is cinematic but intimate, with gentle lens softness and believable environmental detail. The composition balances bodily presence with landscape-driven stillness. Maintain one coherent adult body, natural joints, believable balance and realistic proportions.
El archivo descargable krvulva_traditional_korea_1000.txt contiene 1,000 prompts que abarcan vestimenta tradicional coreana, interiores, paisajes, talleres, templos, costas, luz estacional y pruebas de cámara centradas en anatomía. Son material de prueba, no una promesa de que cada seed se comportará como ciudadano ejemplar.
Esta publicación coreana coloca la quinta y última pieza de la serie de anatomía y realismo humano: occidental, japonesa, china, coreana y masculina. Ya he pasado suficiente tiempo comparando seeds, geometría pélvica, luz ambiental y opiniones privadas de la gravedad como para merecer un breve período de desempleo visual voluntario. Los modelos pueden seguir generando; yo salgo afuera. Si tiene una idea verdaderamente nueva, un informe de fallo útil o una dirección que valga la pena explorar después del descanso, puede contactarme mediante el repositorio. Las buenas ideas siguen siendo interesantes incluso cuando el autor se niega temporalmente a inspeccionar nada con un zoom del 800 %.
Los archivos descargables reference.py y reference_low_vram.py cargan Qwen-Image-2512 en BF16, fusionan el adaptador Lightning comunitario y aplican este LoRA coreano sin activador. Para la versión estándar se recomienda una GPU CUDA con mucha memoria.
pip install -U diffusers transformers accelerate peft safetensors
<details>
<summary><strong>▶ Mostrar el código Python estándar</strong></summary>
"""Qwen Image 2512 + Korean Female Anatomy Realism LoRA v1."""
import math
import torch
from diffusers import DiffusionPipeline, FlowMatchEulerDiscreteScheduler
BASE_MODEL_ID = "Qwen/Qwen-Image-2512"
LIGHTNING_MODEL_ID = "lightx2v/Qwen-Image-2512-Lightning"
LIGHTNING_WEIGHT_NAME = (
"Qwen-Image-2512-Lightning-4steps-V1.0-bf16.safetensors"
)
LORA_MODEL_ID = (
"perpetual3x/Qwen-Image-Korean-Female-Anatomy-Realism-NSFW-LoRA-v1"
)
LORA_WEIGHT_NAME = "krvulva.safetensors"
DEVICE = "cuda"
DTYPE = torch.bfloat16
LORA_SCALE = 0.9
scheduler_config = {
"base_image_seq_len": 256,
"base_shift": math.log(3),
"invert_sigmas": False,
"max_image_seq_len": 8192,
"max_shift": math.log(3),
"num_train_timesteps": 1000,
"shift": 1.0,
"shift_terminal": None,
"stochastic_sampling": False,
"time_shift_type": "exponential",
"use_beta_sigmas": False,
"use_dynamic_shifting": True,
"use_exponential_sigmas": False,
"use_karras_sigmas": False,
}
scheduler = FlowMatchEulerDiscreteScheduler.from_config(scheduler_config)
pipe = DiffusionPipeline.from_pretrained(
BASE_MODEL_ID,
scheduler=scheduler,
torch_dtype=DTYPE,
).to(DEVICE)
pipe.load_lora_weights(
LIGHTNING_MODEL_ID,
weight_name=LIGHTNING_WEIGHT_NAME,
adapter_name="lightning",
)
pipe.fuse_lora(adapter_names=["lightning"], lora_scale=1.0)
pipe.unload_lora_weights()
pipe.load_lora_weights(
LORA_MODEL_ID,
weight_name=LORA_WEIGHT_NAME,
adapter_name="korean_anatomy_v1",
)
pipe.set_adapters(["korean_anatomy_v1"], adapter_weights=[LORA_SCALE])
prompt = (
"A realistic photograph of a clearly adult Korean woman, coherent "
"external female anatomy, natural pelvic structure, unretouched skin "
"texture, physically plausible pose, soft natural light"
)
negative_prompt = (
"child, minor, underage, youthful, low quality, blurry, distorted anatomy, "
"extra limbs, duplicated body parts, fused anatomy, watermark, logo, text"
)
generator = torch.Generator(device=DEVICE).manual_seed(42)
image = pipe(
prompt=prompt,
negative_prompt=negative_prompt,
width=928,
height=1328,
num_inference_steps=8,
true_cfg_scale=1.0,
generator=generator,
).images[0]
image.save("korean_female_anatomy_realism_v1.png")
</details>
<details>
<summary><strong>▶ Mostrar el código Python para poca VRAM</strong></summary>
"""Low-VRAM Qwen Image 2512 + Korean Female Anatomy Realism LoRA v1."""
import gc
import math
import torch
from diffusers import DiffusionPipeline, FlowMatchEulerDiscreteScheduler
BASE_MODEL_ID = "Qwen/Qwen-Image-2512"
LIGHTNING_MODEL_ID = "lightx2v/Qwen-Image-2512-Lightning"
LIGHTNING_WEIGHT_NAME = (
"Qwen-Image-2512-Lightning-4steps-V1.0-bf16.safetensors"
)
LORA_MODEL_ID = (
"perpetual3x/Qwen-Image-Korean-Female-Anatomy-Realism-NSFW-LoRA-v1"
)
LORA_WEIGHT_NAME = "krvulva.safetensors"
DTYPE = torch.bfloat16
LORA_SCALE = 0.9
scheduler_config = {
"base_image_seq_len": 256,
"base_shift": math.log(3),
"invert_sigmas": False,
"max_image_seq_len": 8192,
"max_shift": math.log(3),
"num_train_timesteps": 1000,
"shift": 1.0,
"shift_terminal": None,
"stochastic_sampling": False,
"time_shift_type": "exponential",
"use_beta_sigmas": False,
"use_dynamic_shifting": True,
"use_exponential_sigmas": False,
"use_karras_sigmas": False,
}
gc.collect()
if torch.cuda.is_available():
torch.cuda.empty_cache()
scheduler = FlowMatchEulerDiscreteScheduler.from_config(scheduler_config)
pipe = DiffusionPipeline.from_pretrained(
BASE_MODEL_ID,
scheduler=scheduler,
torch_dtype=DTYPE,
low_cpu_mem_usage=True,
)
pipe.load_lora_weights(
LIGHTNING_MODEL_ID,
weight_name=LIGHTNING_WEIGHT_NAME,
adapter_name="lightning",
)
pipe.fuse_lora(adapter_names=["lightning"], lora_scale=1.0)
pipe.unload_lora_weights()
pipe.load_lora_weights(
LORA_MODEL_ID,
weight_name=LORA_WEIGHT_NAME,
adapter_name="korean_anatomy_v1",
)
pipe.set_adapters(["korean_anatomy_v1"], adapter_weights=[LORA_SCALE])
pipe.enable_sequential_cpu_offload()
pipe.vae.enable_tiling()
pipe.vae.enable_slicing()
prompt = (
"A realistic photograph of a clearly adult Korean woman, coherent "
"external female anatomy, natural pelvic structure, unretouched skin "
"texture, physically plausible pose, soft natural light"
)
negative_prompt = (
"child, minor, underage, youthful, low quality, blurry, distorted anatomy, "
"extra limbs, duplicated body parts, fused anatomy, watermark, logo, text"
)
generator = torch.Generator(device="cpu").manual_seed(42)
image = pipe(
prompt=prompt,
negative_prompt=negative_prompt,
width=768,
height=1024,
num_inference_steps=8,
true_cfg_scale=1.0,
generator=generator,
).images[0]
image.save("korean_female_anatomy_realism_v1_low_vram.png")
</details>
Este modelo está destinado exclusivamente a trabajo creativo adulto, legal y consensuado. No lo utilice para sexualizar menores o personas de apariencia juvenil, crear contenido no consentido o explotador, representar a una persona real identificable sin permiso, acosar o difamar, producir contenido ilegal ni eludir las normas de una plataforma.
Los usuarios son responsables de garantizar que sus prompts, materiales de origen y contenidos generados cumplan las leyes aplicables y las políticas de las plataformas.
Este LoRA solo puede utilizarse con fines personales y no comerciales. Consulte el archivo LICENSE del repositorio para conocer las condiciones completas.
Para consultas sobre licencias comerciales, comentarios técnicos útiles o ideas de proyecto verdaderamente nuevas, contacte al propietario del repositorio.
</details>Adult content warning: This repository contains explicit AI-generated imagery and is intended only for adults. Every person described in the examples and reference prompts is an adult.

This is the final release in my anatomy-and-human-realism series. Western Europe and America opened the wider visual chapter; Japan and China followed with regional female-anatomy studies; the male-anatomy project supplied the half of humanity that image models too often reduce to either polished marble or improvised plumbing. Korea now locks the fifth and final piece into place. It is not last because it matters least. It is last because every puzzle needs one piece that makes the picture suddenly read as complete - and after examining this much generated anatomy, both the author and the Blackwell GPU have earned the right to stop pretending that topology is a leisure activity.
The series ends here deliberately. Individual models may continue to improve, but the planned set is complete: Western, Japanese, Chinese, Korean and male anatomy now occupy the same finished picture. For a while, I plan to rest my eyes, go outside, and look at landscapes in which nobody needs a seed number, an adapter weight, or a detailed discussion of tissue continuity.
Korean visual culture moves easily between hanok courtyards, celadon workshops, coastal wetlands, mountain temples, rain-dark stone, modern interiors, and dense urban light. A model capable of rendering all of that should not reach the human body and suddenly replace observation with a smooth, vaguely confident placeholder.
This LoRA focuses on realistic adult East Asian anatomy, with its strongest training emphasis on Korean women and Korean photographic context. That describes the dataset and its visual distribution—not a claim that millions of people share one face, one body, or one factory configuration.
Beauty systems are very good at polishing surfaces. Anatomy asks a less glamorous question: does everything remain connected, proportioned, and physically plausible after the lighting, pose, gravity, and camera angle have finished negotiating? Skin texture, folds, compression, body hair, and ordinary asymmetry are not rendering failures. They are evidence that the body was allowed to remain a body.
This model therefore aims for believable external anatomy rather than cosmetic perfection. It is trained to preserve the visual continuity among the vulva, clitoral region, perineum, anus, pelvis, and surrounding tissue across both direct anatomical studies and wider environmental compositions.
This is an anatomy-and-realism LoRA. It can work alone when the prompt supplies the adult subject, scene, and composition, or it can act as a specialist beside another NSFW LoRA. When stacking adapters, let the primary LoRA establish identity or style and use this model only as strongly as needed to restore anatomical coherence.
| Item | Value |
|---|---|
| Base model | Qwen Image 2512 |
| Training images | 5,822 |
| GPU | NVIDIA RTX PRO 6000 Blackwell / 96 GB |
| Rank | 32 |
| Batch size | 8 |
| Training steps | 10,000 |
| Trigger | None - natural language only |
0.8-1.2, then adjust for the pose, camera distance, seed, and any other active LoRA0.8, preserve the main LoRA's identity and visual language first, then raise this model only until the anatomical correction becomes stableThe training set is considerably broader than the earlier regional releases, but 5,822 images still do not repeal geometry, gravity, or bad seeds.
All examples depict clearly adult women. The cover prompt is not included. The following eight test cases include their prompts so the model can be evaluated rather than merely admired.
| Generated example |
|---|
![]() |
Prompt
a clearly adult Korean woman age 49 with a slender mature build, a white hemp hanbok robe hanging open and secured behind the hips so no fabric crosses the pelvis, leaning forward with both hands on a low soban table while both feet remain grounded in a pottery workshop in Icheon, surrounded by celadon vessels, clay tools and kiln light. a ground-level rear view looks gently upward beneath the pelvis without hiding the surrounding scene; the anal opening is genuinely readable rather than implied by the gluteal cleft, and the perineum continues coherently toward the vulva. summer sunlight broken by leaves and moving clouds, natural editorial photography with precise anatomy and gentle depth of field; a traditional embroidered cushion and bojagi wrapping cloth add controlled color. Accurate clearly adult proportions, realistic hands and feet, no additional people, no text, no logo, no watermark.
| Generated example |
|---|
![]() |
Prompt
a clearly adult Korean woman age 43 with a compact muscular build, a crimson silk chima untied and pooled around her feet, its matching jeogori worn open, lying on her left side with the lower leg extended and the upper leg supported on a cushion near the reeds of Suncheon Bay, with distant wetlands, birds and a broad luminous sky. a wide lateral composition shows the full body line and cultural setting without obstruction; the vulva, clitoral hood and perineum remain coherent in profile with realistic labial compression and pubic attachment. gentle overcast daylight preserving realistic skin texture, contemporary fine-art realism with authentic material textures and soft asymmetry; a small vase of plum blossoms echoes the season without distracting from the figure. Accurate clearly adult proportions, realistic hands and feet, no additional people, no text, no logo, no watermark.
| Generated example |
|---|
![]() |
Prompt
This intimate documentary-style image records an adult nude Korean woman with realistic proportions and natural skin texture resting on both hands on a long wooden table while turning toward an unseen sound inside a tiled winter garden with citrus trees and a long iron bench. Her softly tousled hair reinforces the unpolished mood. The viewpoint is frontal but slightly off-center, leaving room for reflected light and negative space. North-window light falls softly across the body and leaves the room in gentle shadow. The visual language is cinematic but intimate, with gentle lens softness and believable environmental detail. The image feels intimate without becoming glossy or theatrical. Avoid plastic skin, exaggerated symmetry or fashion retouching; retain natural variation and small imperfections.
| Generated example |
|---|
![]() |
Prompt
This poetic environmental portrait features an adult nude Korean woman with a natural, unretouched appearance resting on both hands on a long wooden table while turning toward an unseen sound inside a faded hotel room with patterned wallpaper and a tall mirror. Her hair is gathered away from her face, leaving the gesture clearly readable. The camera holds a natural eye-level perspective with no exaggerated foreshortening. Backlight through linen curtains outlines the figure while reflected fill retains detail. Its palette is desaturated and painterly, with gentle contrast and no commercial retouching. The atmosphere suggests solitude and an unforced connection with the surrounding place. Preserve full-body continuity, natural hands and feet, plausible weight distribution and understated expression.
| Generated example |
|---|
![]() |
Prompt
A restrained editorial figure study depicts an adult nude Korean woman integrated quietly into the surrounding environment resting on a low stone wall with both feet touching the ground and her torso turned toward the horizon at the edge of an abandoned orchard overlooking a wide agricultural plain. A few damp strands of hair rest naturally against her face and shoulders. A broad side-on composition emphasizes rhythm, balance and environmental texture. Open shade beneath the trees provides even illumination while the distance remains luminous. The print-like finish preserves natural asymmetry, folds and minor real-world imperfections. The emotional tone is reflective and slightly melancholic. Maintain one coherent adult body, natural joints, believable balance and realistic proportions.
| Generated example |
|---|
![]() |
Prompt
The image presents an adult nude Korean woman whose presence feels candid and unperformed lying across a smooth boulder warmed by the remaining daylight on an orchard path covered with fallen blossoms after a spring shower. Her softly tousled hair reinforces the unpolished mood. A distant medium-format composition lets weather and place occupy most of the frame. Soft overcast daylight wraps the figure without hard highlights. The result resembles restrained 35 mm color-negative film, with fine grain, muted saturation and honest texture. The gesture reads as a private moment observed with care and distance. Keep the gesture physically achievable and the relationship between body, gravity and support convincing.
| Generated example |
|---|
![]() |
Prompt
This poetic environmental portrait features an adult nude Korean woman treated as a classical figure study reclining on a long bench beneath tall windows and damaged plaster inside an abandoned winter pavilion with cracked tiles and fogged glass. Her loose hair follows the turn of her head. The camera looks through foreground leaves or fabric, adding a restrained sense of depth. Pale winter light enters through damaged glass and settles quietly across the architecture. The print-like finish preserves natural asymmetry, folds and minor real-world imperfections. The environment and figure share equal visual importance, producing a calm sense of scale. Maintain one coherent adult body, natural joints, believable balance and realistic proportions.
| Generated example |
|---|
![]() |
Prompt
A museum-minded portrait observes an adult nude Korean woman integrated quietly into the surrounding environment reclining on a long bench beneath tall windows and damaged plaster beneath a concrete colonnade after rain, with reflections across the floor. Her hair is tied back in a simple, practical knot. A waist-level side view follows the natural line of the gesture. Late-day light catches the upper walls while reflected fill remains soft below. The visual language is cinematic but intimate, with gentle lens softness and believable environmental detail. The composition balances bodily presence with landscape-driven stillness. Maintain one coherent adult body, natural joints, believable balance and realistic proportions.
The downloadable krvulva_traditional_korea_1000.txt contains 1,000 prompts spanning traditional Korean clothing, interiors, landscapes, workshops, temples, coastlines, seasonal light, and anatomy-focused camera tests. They are test material, not a promise that every seed will behave like a model citizen.
This Korean release places the fifth and final piece into the anatomy-and-human-realism series: Western, Japanese, Chinese, Korean and male. I have now spent enough time comparing seeds, pelvis geometry, environmental light, and the private opinions of gravity to qualify for a short period of deliberate visual unemployment. The models can continue generating; I am going outside. If you have a genuinely new idea, a useful failure report, or a direction worth exploring after the rest period, you are welcome to contact me through the repository. Good ideas remain interesting even when the author is temporarily refusing to inspect anything at 800% zoom.
The downloadable reference.py and reference_low_vram.py load Qwen-Image-2512 in BF16, fuse the community Lightning adapter, and apply this triggerless Korean anatomy-realism LoRA. A high-memory CUDA GPU is recommended for the standard version.
pip install -U diffusers transformers accelerate peft safetensors
<details>
<summary><strong>▶ Show Standard Python Reference</strong></summary>
"""Qwen Image 2512 + Korean Female Anatomy Realism LoRA v1."""
import math
import torch
from diffusers import DiffusionPipeline, FlowMatchEulerDiscreteScheduler
BASE_MODEL_ID = "Qwen/Qwen-Image-2512"
LIGHTNING_MODEL_ID = "lightx2v/Qwen-Image-2512-Lightning"
LIGHTNING_WEIGHT_NAME = (
"Qwen-Image-2512-Lightning-4steps-V1.0-bf16.safetensors"
)
LORA_MODEL_ID = (
"perpetual3x/Qwen-Image-Korean-Female-Anatomy-Realism-NSFW-LoRA-v1"
)
LORA_WEIGHT_NAME = "krvulva.safetensors"
DEVICE = "cuda"
DTYPE = torch.bfloat16
LORA_SCALE = 0.9
scheduler_config = {
"base_image_seq_len": 256,
"base_shift": math.log(3),
"invert_sigmas": False,
"max_image_seq_len": 8192,
"max_shift": math.log(3),
"num_train_timesteps": 1000,
"shift": 1.0,
"shift_terminal": None,
"stochastic_sampling": False,
"time_shift_type": "exponential",
"use_beta_sigmas": False,
"use_dynamic_shifting": True,
"use_exponential_sigmas": False,
"use_karras_sigmas": False,
}
scheduler = FlowMatchEulerDiscreteScheduler.from_config(scheduler_config)
pipe = DiffusionPipeline.from_pretrained(
BASE_MODEL_ID,
scheduler=scheduler,
torch_dtype=DTYPE,
).to(DEVICE)
pipe.load_lora_weights(
LIGHTNING_MODEL_ID,
weight_name=LIGHTNING_WEIGHT_NAME,
adapter_name="lightning",
)
pipe.fuse_lora(adapter_names=["lightning"], lora_scale=1.0)
pipe.unload_lora_weights()
pipe.load_lora_weights(
LORA_MODEL_ID,
weight_name=LORA_WEIGHT_NAME,
adapter_name="korean_anatomy_v1",
)
pipe.set_adapters(["korean_anatomy_v1"], adapter_weights=[LORA_SCALE])
# Natural language only: no trigger word is required.
prompt = (
"A realistic photograph of a clearly adult Korean woman, coherent "
"external female anatomy, natural pelvic structure, unretouched skin "
"texture, physically plausible pose, soft natural light"
)
negative_prompt = (
"child, minor, underage, youthful, low quality, blurry, distorted anatomy, "
"extra limbs, duplicated body parts, fused anatomy, watermark, logo, text"
)
generator = torch.Generator(device=DEVICE).manual_seed(42)
image = pipe(
prompt=prompt,
negative_prompt=negative_prompt,
width=928,
height=1328,
num_inference_steps=8,
true_cfg_scale=1.0,
generator=generator,
).images[0]
image.save("korean_female_anatomy_realism_v1.png")
</details>
<details>
<summary><strong>▶ Show Low-VRAM Python Reference</strong></summary>
"""Low-VRAM Qwen Image 2512 + Korean Female Anatomy Realism LoRA v1."""
import gc
import math
import torch
from diffusers import DiffusionPipeline, FlowMatchEulerDiscreteScheduler
BASE_MODEL_ID = "Qwen/Qwen-Image-2512"
LIGHTNING_MODEL_ID = "lightx2v/Qwen-Image-2512-Lightning"
LIGHTNING_WEIGHT_NAME = (
"Qwen-Image-2512-Lightning-4steps-V1.0-bf16.safetensors"
)
LORA_MODEL_ID = (
"perpetual3x/Qwen-Image-Korean-Female-Anatomy-Realism-NSFW-LoRA-v1"
)
LORA_WEIGHT_NAME = "krvulva.safetensors"
DTYPE = torch.bfloat16
LORA_SCALE = 0.9
scheduler_config = {
"base_image_seq_len": 256,
"base_shift": math.log(3),
"invert_sigmas": False,
"max_image_seq_len": 8192,
"max_shift": math.log(3),
"num_train_timesteps": 1000,
"shift": 1.0,
"shift_terminal": None,
"stochastic_sampling": False,
"time_shift_type": "exponential",
"use_beta_sigmas": False,
"use_dynamic_shifting": True,
"use_exponential_sigmas": False,
"use_karras_sigmas": False,
}
gc.collect()
if torch.cuda.is_available():
torch.cuda.empty_cache()
scheduler = FlowMatchEulerDiscreteScheduler.from_config(scheduler_config)
pipe = DiffusionPipeline.from_pretrained(
BASE_MODEL_ID,
scheduler=scheduler,
torch_dtype=DTYPE,
low_cpu_mem_usage=True,
)
pipe.load_lora_weights(
LIGHTNING_MODEL_ID,
weight_name=LIGHTNING_WEIGHT_NAME,
adapter_name="lightning",
)
pipe.fuse_lora(adapter_names=["lightning"], lora_scale=1.0)
pipe.unload_lora_weights()
pipe.load_lora_weights(
LORA_MODEL_ID,
weight_name=LORA_WEIGHT_NAME,
adapter_name="korean_anatomy_v1",
)
pipe.set_adapters(["korean_anatomy_v1"], adapter_weights=[LORA_SCALE])
pipe.enable_sequential_cpu_offload()
pipe.vae.enable_tiling()
pipe.vae.enable_slicing()
# Natural language only: no trigger word is required.
prompt = (
"A realistic photograph of a clearly adult Korean woman, coherent "
"external female anatomy, natural pelvic structure, unretouched skin "
"texture, physically plausible pose, soft natural light"
)
negative_prompt = (
"child, minor, underage, youthful, low quality, blurry, distorted anatomy, "
"extra limbs, duplicated body parts, fused anatomy, watermark, logo, text"
)
generator = torch.Generator(device="cpu").manual_seed(42)
image = pipe(
prompt=prompt,
negative_prompt=negative_prompt,
width=768,
height=1024,
num_inference_steps=8,
true_cfg_scale=1.0,
generator=generator,
).images[0]
image.save("korean_female_anatomy_realism_v1_low_vram.png")
</details>
This model is intended exclusively for lawful, consensual, adult creative work. Do not use it to:
Users are responsible for ensuring that their prompts, source material, and generated content comply with applicable laws and platform policies.
This LoRA may be used for personal and non-commercial purposes only. See the repository LICENSE file for the complete terms.
For commercial licensing inquiries, useful technical feedback, or genuinely new project ideas, please contact the repository owner.