🎨 Alma SKILL 通过本地 ComfyUI API 生成漫画/插画图片

✨ 特色

• 通过ComfyUI API实现文本转图像生成
• 多种艺术风格支持LoRA：
• 🖌️ 北条司（北条司）- 80年代漫画风格，黑白
• 🎨 漆原智志 - 90年代动漫风格，色彩丰富
• 字符一致性 - 预定义字符，外观固定
• 快速且高分辨率模式 - 快速预览或制作质量
• 放大支持 - 用Ultimate SD放大增强生成图像

📋 先修条件

• 本地或服务器上运行的ComfyUI
• Alma AI 助手
• Z-图像模型：z_image_turbo_bf16.safetensors
• 必需的ComfyUI节点：
• UltimateSDUpscale（用于放大）
• LoadImageFromHttpURL（用于从 URL 升频）

可选LoRA模型

风格转移的话，你需要我Civitai里的LoRA模型：

• 北条司流派：从Civitai下载
• 漆原聪风格：从Civitai下载

🚀 安装

1.克隆仓库

2.安装到 Alma 技能目录

3.配置ComfyUI服务器地址

编辑每个Python脚本并修改SERVER变量：

4.

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/