中文 ｜ English

Project source code address：

Qwen Text2Image & Image2Image Dify Plugin

📖 Project Overview

This is a comprehensive Dify plugin based on ModelScope Qwen-Image models that supports both text-to-image generation and image-to-image editing. Generate high-quality images from text descriptions or edit existing images with text prompts. The plugin uses asynchronous task processing to ensure stable and reliable image generation.

✨ Key Features

• 🎨 High-Quality Image Generation: Supports advanced Qwen-Image-2512 and other models
• ✏️ Image Editing: Edit existing images with text prompts using Qwen-Image-Edit-2511, Qwen-Image-Edit-2509 and other models
• 📐 Custom Image Size Support: Flexible image dimensions with custom size configuration (WxH format)
• 🖼️ Automatic Size Detection: Image2Image tool automatically detects input image dimensions as default
• ⚡ Asynchronous Processing: Uses task submission + polling async mode to avoid timeouts
• 🔄 Real-time Feedback: Provides detailed generation progress and status information
• 🛡️ Error Handling: Comprehensive exception handling with user-friendly error messages
• 🌐 Bilingual Support: Supports both English and Chinese interface and messages

🏗️ Project Architecture

qwen_text2image_plugin/
├── manifest.yaml              # Plugin manifest file
├── main.py                   # Plugin entry point
├── requirements.txt          # Python dependencies
├── .env.example             # Environment variables example
├── README.md                # Project documentation
├── icon.svg                 # Plugin icon
├── provider/                # Service provider configuration
│   ├── __init__.py
│   ├── modelscope.yaml      # ModelScope provider config
│   └── modelscope_provider.py
└── tools/                   # Tool implementation
    ├── __init__.py
    ├── text2image.yaml      # Text-to-image tool config
    └── text2image.py        # Text-to-image tool implementation

🚀 Quick Start

1. Get ModelScope API Key

1. Visit ModelScope Official Website
2. Register and login to your account
3. Go to My Access Token page
4. Create a new API Key (format: ms-xxxxxx)

2. Install Dependencies

pip install -r requirements.txt

3. Configure Environment

Copy .env.example to .env and configure the parameters:

cp .env.example .env

4. Install Plugin in Dify

1. Upload the plugin folder to Dify plugin directory
2. Enable the plugin in Dify management interface
3. Configure ModelScope API Key

🔧 Usage

Basic Usage

1. Add "Qwen Text2Image" tool in Dify workflow

2. Configure ModelScope API Key

   

   

3. Input image description prompt

4. Select model (default: Qwen-Image)

5. Run the tool to generate image

   Workflow DSL example:

   app:
     description: ''
     icon: 🤖
     icon_background: '#FFEAD5'
     mode: advanced-chat
     name: Custom Text2Image Plugin Test - Chatflow
     use_icon_as_answer_icon: false
   dependencies:
   - current_identifier: null
     type: package
     value:
       plugin_unique_identifier: wwwzhouhui/qwen_text2image:0.0.1@18eb2a22be7173a6bd806402b1748b3d7e9967acd87e1b4c5a6b794fa08fca0c
   kind: app
   version: 0.3.0
   workflow:
     conversation_variables: []
     environment_variables: []
     features:
       file_upload:
         enabled: false
       opening_statement: ''
       retriever_resource:
         enabled: true
       sensitive_word_avoidance:
         enabled: false
       speech_to_text:
         enabled: false
       suggested_questions: []
       suggested_questions_after_answer:
         enabled: false
       text_to_speech:
         enabled: false
     graph:
       edges:
       - data:
           isInIteration: false
           isInLoop: false
           sourceType: start
           targetType: tool
         id: 1755656337314-source-1755657278812-target
         source: '1755656337314'
         sourceHandle: source
         target: '1755657278812'
         targetHandle: target
         type: custom
         zIndex: 0
       - data:
           isInLoop: false
           sourceType: tool
           targetType: answer
         id: 1755657278812-source-answer-target
         source: '1755657278812'
         sourceHandle: source
         target: answer
         targetHandle: target
         type: custom
         zIndex: 0
       nodes:
       - data:
           desc: ''
           selected: false
           title: Start
           type: start
           variables: []
         height: 53
         id: '1755656337314'
         position:
           x: 80
           y: 282
         positionAbsolute:
           x: 80
           y: 282
         selected: false
         sourcePosition: right
         targetPosition: left
         type: custom
         width: 244
       - data:
           answer: '{{#1755657278812.text#}}
   
             {{#1755657278812.files#}}
   
             '
           desc: ''
           selected: false
           title: Direct Reply
           type: answer
           variables: []
         height: 123
         id: answer
         position:
           x: 740
           y: 282
         positionAbsolute:
           x: 740
           y: 282
         selected: true
         sourcePosition: right
         targetPosition: left
         type: custom
         width: 244
       - data:
           desc: ''
           is_team_authorization: true
           output_schema: null
           paramSchemas:
           - auto_generate: null
             default: null
             form: llm
             human_description:
               en_US: The text prompt to generate image from. Describe what you want
                 to see in the image in detail. For example "A golden cat sitting on
                 a red sofa in a cozy living room".
             label:
               en_US: Prompt
             llm_description: Text prompt that describes the desired image content in
               detail. The more specific and descriptive, the better the generated image
               quality.
             max: null
             min: null
             name: prompt
             options: []
             placeholder: null
             precision: null
             required: true
             scope: null
             template: null
             type: string
           - auto_generate: null
             default: Qwen/Qwen-Image
             form: form
             human_description:
               en_US: The AI model to use for image generation. Qwen-Image is the default
                 and recommended model.
             label:
               en_US: Model
             llm_description: ''
             max: null
             min: null
             name: model
             options:
             - icon: ''
               label:
                 en_US: Qwen-Image (Recommended)
               value: Qwen/Qwen-Image
             placeholder: null
             precision: null
             required: false
             scope: null
             template: null
             type: select
           params:
             model: ''
             prompt: ''
           provider_id: wwwzhouhui/qwen_text2image/modelscope
           provider_name: wwwzhouhui/qwen_text2image/modelscope
           provider_type: builtin
           selected: false
           title: Text to Image
           tool_configurations:
             model:
               type: constant
               value: Qwen/Qwen-Image
           tool_description: Generate high-quality images from text prompts using ModelScope
             Qwen-Image AI model. Support various image styles and detailed descriptions.
           tool_label: Text to Image
           tool_name: text2image
           tool_parameters:
             prompt:
               type: mixed
               value: '{{#sys.query#}}'
           type: tool
           version: '2'
         height: 121
         id: '1755657278812'
         position:
           x: 384
           y: 282
         positionAbsolute:
           x: 384
           y: 282
         selected: false
         sourcePosition: right
         targetPosition: left
         type: custom
         width: 244
       viewport:
         x: 34
         y: 87.5
         zoom: 1
   

   

Prompt Suggestions

For best image generation results, we recommend:

• Detailed Description: Provide specific information about scene, objects, colors, styles, etc.
• Clear Expression: Use concise and clear language for description
• Style Specification: You can specify artistic styles like "oil painting style", "cartoon style", etc.

Example prompt:

A golden cat sitting on a red sofa in a cozy living room, with warm sunlight streaming through the window, creating a cozy home atmosphere

⚙️ Technical Implementation

Core Workflow

1. Task Submission: Submit asynchronous image generation task to ModelScope API
2. Status Polling: Query task status every 5 seconds, wait up to 5 minutes
3. Image Download: Download generated image after task completion
4. Format Conversion: Use PIL to convert image to PNG format and return

API Call Pattern

# 1. Submit task
POST /v1/images/generations
Headers: X-ModelScope-Async-Mode: true

# 2. Query status
GET /v1/tasks/{task_id}
Headers: X-ModelScope-Task-Type: image_generation

# 3. Download image
GET {image_url}

🔍 Troubleshooting

Common Issues

1. Invalid API Key

   • Check if API Key format starts with ms-
   • Confirm API Key is valid and not expired

2. Generation Timeout

   • Check if network connection is normal
   • Try simplifying prompt description
   • Retry later

3. Image Download Failed

   • Check network connection
   • Confirm firewall settings allow access to ModelScope domains

Error Codes

• 401: Invalid or unauthorized API Key
• 429: API call rate limit exceeded
• 500: Internal server error

📋 Development Standards

This plugin strictly follows the Dify text-to-image plugin development standards defined in CLAUDE2.md:

• ✅ Asynchronous task processing mode
• ✅ Complete error handling mechanism
• ✅ Real-time progress feedback
• ✅ Bilingual support (English/Chinese)
• ✅ Standard ModelScope API calls

🤝 Contributing

Welcome to submit Issues and Pull Requests to improve this plugin!

📄 License

This project is licensed under the MIT License.

🔗 Related Links

• ModelScope Official Website
• Qwen-Image Model
• Dify Official Documentation

📦 Release Notes

0.0.4

• Fixed Image2Image Functionality: Resolved issue where ModelScope server couldn't access Dify internal image URLs causing task failures
• Added Temporary Image Hosting: Integrated litterbox.catbox.moe temporary image hosting service to automatically upload images and obtain publicly accessible URLs
• Improved Image Processing:
  • Automatically handles RGBA, LA, P and other color modes, converting to RGB format
  • Supports image URLs from various sources (including Dify internal addresses, intranet addresses, etc.)
• Increased Network Timeout:
  • API submission request timeout: 300 seconds
  • Task status polling timeout: 120 seconds
  • Image hosting upload timeout: 120 seconds
• Updated Default Model: Image2Image default model updated to Qwen/Qwen-Image-Edit-2511
• Improved Error Handling: Optimized error message extraction logic, providing more detailed debugging information

0.0.3

• Enhanced Custom Image Size Support: Both Text2Image and Image2Image tools now support flexible custom image dimensions
• Automatic Size Detection: Image2Image tool automatically detects and uses input image dimensions as default size
• Improved Size Validation: Added comprehensive size format validation with user-friendly error messages
• Better Error Handling: Enhanced error messages for invalid size parameters with automatic fallback
• Code Optimization: Improved parameter handling and validation logic in both tools
• Updated Documentation: Enhanced README with detailed size configuration examples and usage guidelines

0.0.2

• Added Image-to-Image tool (Image2Image) based on ModelScope Qwen-Image-Edit
• New files: tools/image2image.py, tools/image2image.yaml
• Registered the tool in provider/modelscope.yaml and imported in provider/modelscope_provider.py
• Updated manifest.yaml description and labels to reflect both text-to-image and image-to-image
• Updated README docs (EN/ZH)
• Backward compatible; no breaking changes; existing Text2Image workflows are unaffected
• Usage: In Dify, choose the "Image to Image" tool, then provide a prompt and a public image URL

0.0.1

• Initial release with Text2Image tool based on ModelScope Qwen-Image