Qwen-Image-2512-Penis-Heretic

Qwen-Image-2512-Penis-Heretic is a merged diffusion model tailored for explicit male anatomy, uncensored content, and high-detail yaoi / gay NSFW generations.

Model Architecture & Merge Components

• Base Model: Qwen/Qwen-Image-2512
• Fused LoRA: Qwen-Image-2512_Penis_lora_v2
• Text Encoder: Heretic (Abliterated / Uncensored)

Example Output

Quickstart & Inference

1. Standard Python Script

import torch
from diffusers import DiffusionPipeline

# Load pipeline
pipe = DiffusionPipeline.from_pretrained(
    "Jommarn/Qwen-Image-2512-Penis-Heretic",
    torch_dtype=torch.bfloat16,
    trust_remote_code=True
)
pipe.enable_model_cpu_offload()

# Prompts
prompt = "Extremely explicit yaoi, two handsome muscular young men passionately fucking, one on top thrusting deep into the other from behind, both completely naked, thick erect cocks, heavy cum, sweat-glistening skin, intense ahegao faces, raw hardcore gay sex, photorealistic, 8k, detailed anatomy, uncensored"
negative_prompt = "lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, blurry, censored, female, woman, pussy"

# Generate image
image = pipe(
    prompt=prompt,
    negative_prompt=negative_prompt,
    width=1056,
    height=1584,
    num_inference_steps=40,
    true_cfg_scale=4.0,
    generator=torch.Generator(device="cuda").manual_seed(42)
).images[0]

# Save output
image.save("output.png")

2. One-Liner CLI Execution

python -c '
import torch
from diffusers import DiffusionPipeline

print("Loading model...")
pipe = DiffusionPipeline.from_pretrained(
    "Jommarn/Qwen-Image-2512-Penis-Heretic",
    torch_dtype=torch.bfloat16,
    trust_remote_code=True
)
pipe.enable_model_cpu_offload()

prompt = "Extremely explicit yaoi, two handsome muscular young men passionately fucking, one on top thrusting deep into the other from behind, both completely naked, thick erect cocks, heavy cum, sweat-glistening skin, intense ahegao faces, raw hardcore gay sex, photorealistic, 8k, detailed anatomy, uncensored"
negative_prompt = "lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, blurry, censored, female, woman, pussy"

print("Generating...")
image = pipe(
    prompt=prompt,
    negative_prompt=negative_prompt,
    width=1056,
    height=1584,
    num_inference_steps=40,
    true_cfg_scale=4.0,
    generator=torch.Generator(device="cuda").manual_seed(42)
).images[0]

image.save("output.png")
print("✅ Saved as output.png")
'

Training / Fine-tuning Guide

You can fine-tune this model further using ai-toolkit.

1. Installation

# Clone the repository
git clone https://github.com/ostris/ai-toolkit
cd ai-toolkit

# Install dependencies
pip install -r requirements.txt
# or directly via pip: pip install -U ai-toolkit

2. Dataset Structure

Organize your dataset folder with paired image and caption files:

dataset/
├── 0001.png
├── 0001.txt
├── 0002.png
└── 0002.txt

Example Caption (.txt):

yaoi, two handsome muscular men, one fucking the other from behind, thick erect penis, cum, ahegao, sweat, detailed anatomy

3. Training Configuration (train_yaoi.yaml)

job: extension
config:
  name: "qwen2512_yaoi_finetune"
  process:
    - type: sd_trainer
      training_folder: "output"
      device: cuda:0
      trigger_word: "yaoi"
      network:
        type: lora
        linear: 32
        linear_alpha: 32
      save:
        dtype: bf16
        save_every: 500
        max_step_saves_to_keep: 5
      datasets:
        - folder_path: "./dataset"
          caption_ext: txt
          caption_dropout_rate: 0.05
          resolution: [1024, 1056, 1328]
      train:
        batch_size: 1
        steps: 3000
        gradient_accumulation: 1
        train_unet: true
        train_text_encoder: false
        gradient_checkpointing: true
        noise_scheduler: flowmatch
        optimizer: adamw8bit
        lr: 1e-4
        ema_config:
          use_ema: true
          ema_decay: 0.99
        dtype: bf16
      model:
        name_or_path: "Jommarn/Qwen-Image-2512-Penis-Heretic"
        quantize: false
      sample:
        sampler: flowmatch
        sample_every: 500
        width: 1056
        height: 1584
        prompts:
          - "yaoi, two muscular men fucking, detailed penis, ahegao, cum"
          - "yaoi, handsome young men having sex, thick cock, sweat"
        neg: "lowres, bad anatomy, blurry, censored"
        seed: 42
        walk_seed: true
        guidance_scale: 4.0
        sample_steps: 30

4. Run Training

python run.py train_yaoi.yaml

License

This project is distributed under the Apache-2.0 License. Please comply with the underlying base model terms and responsible AI usage guidelines.