ModelScope Z-Image Generator

Generate images using ModelScope's Z-Image series models.

IMPORTANT: Response Format

After generating an image, you MUST:

1. Use the Python script to generate the image
2. Respond with ONLY a simple text message like: "Image saved to: /path/to/output.jpg"
3. DO NOT include the image URL in your response
4. DO NOT try to display or reference the image URL - this will cause errors with non-multimodal models

The user can view the generated image at the file path you provide.

Prerequisites

Get your ModelScope API key from: https://modelscope.cn/my/myaccesstoken

The script will automatically prompt for your API key the first time you use it and offer to save it for future use.

Optional: Manual Configuration

You can also manually configure the API key:

Config file:

mkdir -p ~/.config/modelscope
cat > ~/.config/modelscope/config.json << EOF
{
  "api_key": "ms-your-api-key-here"
}
EOF

Environment variable:

export MODELSCOPE_API_KEY="ms-your-api-key-here"

Quick Start

Use the Python script to generate images:

cd /Users/ningoo/.claude/skills/modelscope-image-generator/scripts
python generate_image.py "A golden cat" output.jpg

Models

Available Z-Image models:

• Tongyi-MAI/Z-Image-Turbo (default, fast generation)
• Tongyi-MAI/Z-Image (balanced quality and speed)
• Tongyi-MAI/Z-Image-Edit (image editing)

Model Selection Rule

• If the user explicitly mentions "Z-Image-Turbo", use Tongyi-MAI/Z-Image-Turbo
• If the user explicitly mentions "Z-Image" (without Turbo), use Tongyi-MAI/Z-Image
• Otherwise, use the default Tongyi-MAI/Z-Image-Turbo

Specify a different model using --model:

python generate_image.py "A golden cat" output.jpg --model "Tongyi-MAI/Z-Image-Turbo"

LoRA Support

Single LoRA:

"loras": "<lora-repo-id>"

Multiple LoRAs (weights must sum to 1.0):

"loras": {"<lora-id1>": 0.6, "<lora-id2>": 0.4}

API Flow

1. Submit generation request with X-ModelScope-Async-Mode: true
2. Receive task_id in response
3. Poll /v1/tasks/{task_id} with X-ModelScope-Task-Type: image_generation
4. Wait for status SUCCEED or FAILED
5. Download image from output_images[0] URL