Vermouth (Chris Vineyard) (Ilustrious XL) (Detective Conan)

ES: "A secret makes a woman, woman." Captura la esencia de la integrante más misteriosa de la Organización de Negro. Este LoRA ha sido entrenado para replicar la elegancia atemporal y la mirada gélida de Vermouth (Sharon/Chris Vineyard).

EN: "A secret makes a woman, woman." Capture the essence of the Black Organization's most mysterious member. This LoRA has been trained to replicate the timeless elegance and ice-cold gaze of Vermouth (Sharon/Chris Vineyard).

Recommended resolutions:

Start with 512 x 512 px or 512 x 768 px (rescale with R-ESRGAN_x4Plus Anime 6B)

Avoid 360 x 640 px!!!

Aspect Ratio Tip: Always use dimensions that are multiples of 64 (e.g., 832x1216 or 1344x768) to avoid generation artifacts and ensure maximum sharpness.

Square (Cuadrado)

1024 x 1024 — Ratio 1:1

Portrait (Vertical)

896 x 1152 — Ratio 7:9

832 x 1216 — Ratio 2:3

768 x 1344 — Ratio 9:16

640 x 1536 — Ratio 9:21

Landscape (Horizontal)

1152 x 896 — Ratio 9:7

1216 x 832 — Ratio 3:2

1344 x 768 — Ratio 16:9

1536 x 640 — Ratio 21:9

trigger words: Vermouth, 1girl, solo, (((tall))), Meitantei_Conan, (mature woman), toned body, slim waist, nsfw, sexy, emerald green eyes, bangs, blonde hair, long hair, red lips, natural elegance

🛠️ Technical Specifications: Vermouth (Chris Vineyard) LoRA

This model was trained using the Kohya_ss framework, focusing on stability and high fidelity to the Detective Conan artistic style.

📊 Training Dataset

• Base Model: SDXL 1.0 (Stable Diffusion XL)

• Dataset Size: 17 high-quality images.

• Repetitions: 10 per image.

• Total Steps: 1,700 steps.

• Training Resolution: 1024 x 1024 px.

⚙️ Architecture & Hyperparameters

• Network Rank (Dimension): 64

• Network Alpha: 32.0 (Provides a stable 0.5 ratio for weight distribution).

• Learning Rate:

  • UNet: 0.0005

  • Text Encoder: 0.00005

• Optimizer: AdamW8bit

• LR Scheduler: Cosine with Restarts.

• Precision: Mixed Precision FP16.

🎨 Advanced Features

• Clip Skip: 2 (Optimized for anime-style inference).

• Noise Offset: 0.03 (Improved contrast for dark/noir scenes, typical of the Black Organization).

• Multires Noise: Enabled (10 iterations), which helps the model capture better details at different scales.

🚀 Recommended Settings for Generation

• LoRA Weight: 0.8 - 1.0

• Sampler: DPM++ 2M Karras or Euler a.

• Steps: 25 - 35 steps.

• CFG Scale: 5.0 - 7.5.

• Clip Skip: 2 (very important!)