Chinese Female NudeShot NSFW LoRA v1

Chinese Female NudeShot NSFW LoRA v1 is a photorealistic LoRA trained from adult Chinese female nude photo-book photography. It is compatible with and tested on both Qwen/Qwen-Image-2512 Text-to-Image and Qwen/Qwen-Image-Edit-2511.

The base model has broad visual knowledge, but its default output does not strongly represent Chinese women or the visual character of Chinese photo-book photography. This LoRA was trained specifically to strengthen that underrepresented area. We believe users will be able to see a clear difference between the East Asian female characteristics learned by this LoRA and the more Western-oriented female appearance often produced by the base model alone.

The model was trained to strengthen:

• realistic photo-book and editorial photography aesthetics
• natural skin texture and photographic detail
• soft indoor, window, studio, and outdoor lighting
• full-body composition and natural camera framing
• authentic room, furniture, and environmental context
• reduced synthetic, plastic, or CGI-like appearance

Important anatomy notice: This LoRA was trained from photographic nude photo-book material. It is not an anatomical or medical model and does not guarantee correct human anatomy. Generated hands, limbs, joints, body proportions, or intimate anatomy may contain errors. Review results carefully and do not use generated images as anatomical learning references.

Content notice: This project focuses on lawful adult artistic nude, photo-book, and editorial photography. It is not intended as pornography or for depicting sexual acts. The repository nevertheless contains explicit adult nudity and is intended only for adults.

✨ Features

• Photorealistic adult Chinese female portraiture
• Real-photo-trained visual characteristics
• Artistic nude and photo-book aesthetics
• Natural skin, lighting, shadows, and depth
• Indoor bedroom, studio, window-light, and outdoor scenes
• Portrait, seated, reclining, and full-body compositions
• Tested with Qwen-Image-2512 T2I
• Tested with Qwen-Image-Edit-2511

🖥️ Training Details

Hardware

• GPU: NVIDIA RTX PRO 6000 Blackwell 96GB

Dataset

• Training material: 16,026 photographic images
• Source style: adult nude photo-book photography
• Dataset period: 2010–2018
• Trigger word: zl_chinanudefemale_v1
• LoRA rank: 32

Training Schedule

• Final checkpoint: Step 15,000
• The second stage resumed from the Step 9,000 checkpoint.

⚠️ Base Model Compatibility

This LoRA is compatible with and tested on:

• Qwen/Qwen-Image-2512 for Text-to-Image generation
• Qwen/Qwen-Image-Edit-2511 for Image Edit workflows

Use Qwen/Qwen-Image-2512 when generating directly from text. Use Qwen/Qwen-Image-Edit-2511 when applying the learned photographic style or appearance during image editing.

Compatibility with other Qwen-Image checkpoints or architectures is not guaranteed. An incompatible base model may produce weak effects, visual artifacts, incorrect colors, or a loading error.

🔍 Scope and Limitations

This LoRA learns visual patterns from a large collection of real photo-book images. Its purpose is to improve photographic character, framing, lighting, skin rendering, and scene atmosphere.

It does not understand the human body as a formal anatomical system. Dataset bias, difficult poses, occlusion, unusual camera angles, and explicit anatomy can still cause:

• incorrect hands, fingers, feet, or limbs
• inconsistent joints or body proportions
• duplicated, missing, or merged body parts
• implausible poses or perspective
• inaccurate intimate anatomy
• unwanted artifacts in clothing or backgrounds

Prompt wording, seed, resolution, LoRA weight, inference steps, and the selected base model all affect the result. Regenerate or edit visible defects when necessary.

🖼️ Examples

The following nine images were generated with this LoRA. All people described in the prompts are adults. The cover image n0.png is not associated with a prompt.

Example 1

Prompt

A nude adult Asian woman with long, dark brown hair is lying on her back on a bed with white sheets. She is positioned with her legs spread apart, her left leg bent at the knee and her right leg straight. Her left arm is raised above her head, gripping the edge of the bed frame, while her right hand rests on her stomach. The woman has a medium build, visible breasts, and natural pubic hair. The background features a beige wall and a wooden headboard with intricate carvings. The lighting is warm, casting soft shadows on the wall and bed. The image is taken from a side angle, capturing the woman's full body and the surrounding environment.

Example 2

Prompt

A nude adult Asian woman with light skin and straight black hair styled in a bun sits on a white ledge. She has a slim build with small breasts. Her legs are crossed at the ankles, and her right hand rests on her knee while her left hand is placed on her cheek. She is looking directly at the camera with a neutral expression. Behind her is a white curtain with vertical stripes, partially obscuring a window. To her left is a potted plant with pink and purple flowers. The floor and ledge are white, creating a bright, clean, minimalist room.

Example 3

Prompt

This is a photograph of a nude adult Asian woman sitting on a bed. She has a fair complexion and is positioned with her legs spread apart and her feet prominently in the foreground. She looks directly at the camera with a neutral expression. Her left hand rests on her left thigh, and her right hand is placed on the bed beside her. The bed has a white ornate metal headboard with intricate scrollwork mounted against a light pink wall. The white bedspread has a subtle floral pattern, and two pillows with pink polka dots and ruffled edges sit behind her. The setting is a feminine, playfully decorated bedroom.

Example 4

Prompt

An adult Asian woman with fair skin is standing nude in a room. She leans against a white radiator with her legs slightly apart and her hands resting on her hips. She has a slender build and looks directly at the camera with a slight smile. Her fingernails are painted with a light-colored polish. A large potted plant with long green leaves fills the foreground. The background shows a window with white blinds, and the dimly lit room suggests nighttime outside. Light-colored wooden planks cover the floor, creating a cozy, intimate setting.

Example 5

Prompt

zl_chinanudefemale_v1, An adult woman sits nude on a brown armchair in a well-lit room. She has a neutral expression and her legs are spread apart. Her right hand rests on her lap, while her left hand is placed on her chest. The room features beige curtains, a wooden door, and a glass coffee table with a small ashtray on it. A trash can is visible in the foreground.

Example 6

Prompt

A nude adult woman with long dark hair is positioned on all fours on a green pool table. Her body is angled slightly toward the camera, and she turns her head back over her shoulder to look directly at the viewer. She has light skin and appears to be of East Asian descent. Several pool balls are scattered across the green felt, including a black eight ball and yellow, red-striped, purple, and orange-striped balls. Large windows with sheer white curtains allow natural light into the room, and a potted plant with large green leaves stands to the left of the frame.

Example 7

Prompt

The image features a topless adult Asian woman standing in a lush green outdoor setting. She wears a flowing red skirt tied around her waist. Her right hand raises part of the skirt fabric while her left hand rests on her hip, and she wears black high-heeled sandals. The background combines dense green foliage with a weathered gray stone wall. Bright daytime lighting and an overexposed patch of sky in the upper-right corner give the scene a natural, untamed atmosphere.

Example 8

Prompt

This is an outdoor photograph taken on a cloudy day in a park-like green space with grass and paved pathways. In the foreground, an adult Asian woman lies on her stomach in the grass, with her legs bent at the knees and her feet raised. She wears a blue-and-white floral dress that is lifted at the back, along with light-colored sandals. Her hair is long and dark. Trees, bushes, lampposts, and signs fill the well-maintained background. The overcast sky gives the scene a muted gray-green tone and a calm atmosphere.

Example 9

Prompt

An adult Asian woman with long, straight black hair stands in front of a large window with sheer white curtains. She wears a white lace blouse that is unbuttoned, revealing her bare chest and torso. Her right hand is placed on her hip, and she wears white lace-trimmed underwear. Bright natural light floods through the window, creating a soft, diffused photographic effect.

📄 Training Prompts

The training prompt list is included with this repository. Each prompt occupies one line, making the file easy to reuse or adapt for training, testing, and sample generation.

Download prompts.txt

🚀 Reference Code — Qwen-Image-2512 T2I

• This example loads the full Qwen-Image-2512 model in BF16.
• It uses Qwen-Image-2512-Lightning 4-step acceleration.
• It does not use CPU offloading or base-model quantization.
• A high-memory CUDA GPU is recommended.

Installation

pip install -U diffusers transformers accelerate peft safetensors

Python

import math

import torch
from diffusers import DiffusionPipeline, FlowMatchEulerDiscreteScheduler


# Model configuration
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/Chinese-Female-NudeShot-NSFW-LoRA-v1"
LORA_WEIGHT_NAME = "chinanude.safetensors"

DEVICE = "cuda"
DTYPE = torch.bfloat16


# Scheduler
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)


# Load Qwen-Image-2512
pipe = DiffusionPipeline.from_pretrained(
    BASE_MODEL_ID,
    scheduler=scheduler,
    torch_dtype=DTYPE,
).to(DEVICE)


# Fuse the 4-step Lightning LoRA
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()


# Load Chinese Female NudeShot LoRA
pipe.load_lora_weights(
    LORA_MODEL_ID,
    weight_name=LORA_WEIGHT_NAME,
    adapter_name="chinanude",
)
pipe.set_adapters(["chinanude"], adapter_weights=[1.0])


# Generate
prompt = (
    "zl_chinanudefemale_v1, an adult Chinese woman in an artistic nude "
    "photo-book portrait, realistic photography, natural skin texture, "
    "soft window light, tasteful editorial composition"
)

negative_prompt = (
    "low quality, blurry, distorted anatomy, malformed hands, extra limbs, "
    "duplicated body parts, plastic skin, CGI, watermark, text"
)

generator = torch.Generator(device=DEVICE).manual_seed(42)

image = pipe(
    prompt=prompt,
    negative_prompt=negative_prompt,
    width=928,
    height=1328,
    num_inference_steps=4,
    true_cfg_scale=1.0,
    generator=generator,
).images[0]

image.save("chinese_female_nudeshot.png")

🧩 Low-VRAM Inference (16–24 GB)

The LoRA itself is relatively small. Most VRAM is consumed by the Qwen-Image-2512 base model and its text encoder. Users with lower-memory NVIDIA GPUs can combine 4-bit quantization, CPU offloading, VAE tiling, batch size 1, and a lower output resolution.

This configuration is intended as a practical starting point for GPUs with approximately 16–24 GB VRAM. Actual memory use and speed depend on the GPU, system RAM, CUDA version, Diffusers version, prompt length, and resolution.

Low-VRAM Installation

pip install -U diffusers transformers accelerate peft safetensors bitsandbytes

Low-VRAM Python

import torch
from diffusers import DiffusionPipeline
from diffusers.quantizers import PipelineQuantizationConfig


BASE_MODEL_ID = "Qwen/Qwen-Image-2512"
LORA_MODEL_ID = "perpetual3x/Chinese-Female-NudeShot-NSFW-LoRA-v1"
LORA_WEIGHT_NAME = "chinanude.safetensors"


# Quantize the Qwen transformer and text encoder to 4-bit NF4.
quant_config = PipelineQuantizationConfig(
    quant_backend="bitsandbytes_4bit",
    quant_kwargs={
        "load_in_4bit": True,
        "bnb_4bit_quant_type": "nf4",
        "bnb_4bit_compute_dtype": torch.bfloat16,
    },
    components_to_quantize=[
        "transformer",
        "text_encoder",
    ],
)


pipe = DiffusionPipeline.from_pretrained(
    BASE_MODEL_ID,
    torch_dtype=torch.bfloat16,
    quantization_config=quant_config,
)

pipe.load_lora_weights(
    LORA_MODEL_ID,
    weight_name=LORA_WEIGHT_NAME,
    adapter_name="chinanude",
)
pipe.set_adapters(
    ["chinanude"],
    adapter_weights=[1.0],
)

# Do not call pipe.to("cuda") when CPU offloading is enabled.
pipe.enable_model_cpu_offload()
pipe.vae.enable_tiling()

prompt = (
    "zl_chinanudefemale_v1, an adult Chinese woman, realistic "
    "photo-book photography, natural skin texture, soft photographic "
    "lighting, authentic details"
)

negative_prompt = (
    "low quality, blurry, distorted anatomy, malformed hands, "
    "extra limbs, duplicated body parts, watermark, text"
)

generator = torch.Generator(device="cuda").manual_seed(42)

image = pipe(
    prompt=prompt,
    negative_prompt=negative_prompt,
    width=768,
    height=1024,
    num_images_per_prompt=1,
    num_inference_steps=20,
    true_cfg_scale=3.0,
    max_sequence_length=256,
    generator=generator,
).images[0]

image.save("chinese_female_nudeshot_low_vram.png")

If this still causes an out-of-memory error, replace:

pipe.enable_model_cpu_offload()

with:

pipe.enable_sequential_cpu_offload()

Sequential CPU offloading uses less GPU memory, but it is substantially slower. Do not move the pipeline to CUDA before enabling either offloading mode.

For additional details, see the official Diffusers documentation for quantization and memory optimization.

🔞 Responsible Use

This model is intended exclusively for lawful, consensual, adult artistic and editorial creative work. It is not intended for pornography or sexual acts.

Do not use this LoRA to:

• create or sexualize minors or youthful-looking persons
• create non-consensual or exploitative sexual content
• depict an identifiable real person without permission
• harass, threaten, exploit, or defame another person
• create illegal content
• bypass the rules of platforms or services

Users are responsible for ensuring that their prompts, source material, and generated content comply with applicable laws and platform policies.

License & Copyright Notice

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, please contact the author.

🔮 V2 Preview

Chinese Female NudeShot NSFW LoRA v2 is currently in development. V2 uses a newly curated dataset composed primarily of material published after 2024. Its main objective is to improve the visual accuracy and consistency of specific anatomical features and organs, especially in close-up and detail-sensitive compositions.

The new dataset is intended to reduce inconsistent shapes, unclear local structure, and other detail errors seen in difficult angles or tightly framed images. These improvements are designed to increase visual fidelity, but V2 will still not be an anatomical or medical model and cannot guarantee perfect human anatomy in every generation.

The following images are early V2 preview results.

V2 Preview 1

V2 Preview 2