ComfyUI Z-Image SKILL

🎨 An Alma SKILL for generating manga/illustration images via local ComfyUI API

✨ Features

• Text-to-Image Generation via ComfyUI API
• Multiple Art Styles with LoRA support:
  • 🖌️ Hojo Tsukasa (北条司) - 80s manga style, black & white
  • 🎨 Satoshi Urushihara (漆原智志) - 90s anime style, colorful
• Character Consistency - Pre-defined character with fixed appearance
• Quick & High-Res Modes - Fast preview or production quality
• Upscaling Support - Enhance generated images with Ultimate SD Upscale

📋 Prerequisites

• ComfyUI running locally or on a server
• Alma AI assistant
• Z-Image model: z_image_turbo_bf16.safetensors
• Required ComfyUI nodes:
  • UltimateSDUpscale (for upscaling)
  • LoadImageFromHttpURL (for upscaling from URL)

Optional LoRA Models

For style transfer, you’ll need LoRA models from my Civitai:

• Hojo Tsukasa Style: Download from Civitai
• Satoshi Urushihara Style: Download from Civitai

🚀 Installation

1. Clone the repository

git clone https://github.com/norsizu/comfyui-zimage-skill.git

2. Install to Alma skills directory

# macOS
cp -r comfyui-zimage-skill ~/.config/alma/skills/

# Or create a symlink
ln -s /path/to/comfyui-zimage-skill ~/.config/alma/skills/comfyui-zimage

3. Configure ComfyUI server address

Edit each Python script and modify the SERVER variable:

# ============ CONFIGURATION - MODIFY THESE ============
SERVER = "localhost:8188"  # Your ComfyUI server address
# ======================================================

4. Configure LoRA paths (if using style transfer)

Update LoRA paths in SKILL.md to match your ComfyUI models directory:

--lora "YOUR_LORA_FOLDER\\z-image-hoko.safetensors"

📁 Project Structure

comfyui-zimage-skill/
├── SKILL.md                    # Main skill document (Alma reads this)
├── generate_riko_quick.py      # Quick generation for Riko character
├── generate_riko_highres.py    # High-res generation for Riko character
├── generate_manga_quick.py     # Quick manga style generation
├── generate_manga_highres.py   # High-res manga style generation
├── upscale_riko.py             # Upscale Riko images (no LoRA)
├── upscale_manga.py            # Upscale manga images (with LoRA)
├── data/
│   ├── riko_character.py       # Character appearance & outfits
│   ├── style_hojo.md           # Hojo style prompt guide
│   ├── style_satoshi.md        # Satoshi style prompt guide
│   └── __init__.py
├── downloads/                  # Generated images saved here
└── README.md

🎯 Usage

In Alma Chat

Once installed, simply ask Alma to generate images:

"Generate a portrait of a girl in Hojo style"
"Draw Riko in school uniform, looking at cherry blossoms"
"Create a cyberpunk warrior in Satoshi style"

Command Line (Direct)

# Quick manga generation with Hojo style
python3 generate_manga_quick.py \
  --lora "Z-image\\z-image-hojo.safetensors" \
  --prompt "Zanshou_kin_Hojo, 1girl, solo, office lady, cityscape, monochrome"

# Character generation (Riko)
python3 generate_riko_quick.py \
  --outfit_type school_uniform \
  --scene "standing by window, soft sunlight, peaceful expression"

# High-resolution version
python3 generate_manga_highres.py \
  --lora "Z-image\\z-image-satoshi.safetensors" \
  --prompt "sato, 1girl, fantasy warrior, armor, sunset, shiny skin"

🎨 Style Comparison

👋 About Me

• Twitter: @norsizu
• Civitai: norsizu

🙏 Acknowledgments

• Alma by @yetone - Amazing AI assistant with SKILL support
• skill-prompt-generator by @huangserva - Inspiration for this project
• ComfyUI - Powerful and modular Stable Diffusion GUI
• Z-Image - Base model for generation

📄 License

MIT License - Feel free to use, modify, and distribute.