A custom node for ComfyUI that provides seamless integration with the Qwen multimodal models from Alibaba Cloud Model Studio. This solution delivers cutting-edge image generation, editing, and vision capabilities directly within ComfyUI.
This is a direct integration with Alibaba Cloud's Model Studio service, not a third-party wrapper or local model implementation. Benefits include:
- Enterprise-Grade Infrastructure: Leverages Alibaba Cloud's battle-tested AI platform serving millions of requests daily
- State-of-the-Art Models: Access to the latest Qwen models (qwen-image, qwen-image-edit, qwen-vl-max, qwen-vl-plus) with continuous updates
- Commercial Licensing: Properly licensed for commercial use through Alibaba Cloud's terms of service
- Scalable Architecture: Handles high-volume workloads with Alibaba Cloud's reliable infrastructure
- Security Compliance: Follows Alibaba Cloud's security best practices with secure API key management
Model Authorization Required: If you're using a non-default workspace or project in Alibaba Cloud, you may need to explicitly authorize access to the models in your DashScope console.
- Generate images from text (T2I)
- Edit existing images based on text instructions (I2I)
- Multi-image editing support (up to 3 images for advanced editing workflows)
- Analyze and describe images using Qwen-VL models
- Configurable parameters: region, seed, resolution, prompt extension, watermark, negative prompts
- All nodes now return the image URL in addition to the image tensor
- Support for both international and mainland China API endpoints
- Support for separate API keys for different regions
- Automatic environment variable loading from config directory
- Industry-standard directory organization (core, config, generators)
- Powered by Alibaba Cloud's advanced Qwen models
-
Clone this repository to your ComfyUI custom nodes directory:
cd ComfyUI/custom_nodes git clone https://github.com/ru4ls/ComfyUI_Qwen_Image.git -
Install the required dependencies:
pip install -r ComfyUI_Qwen_Image/requirements.txt
- Visit Alibaba Cloud Model Studio to get your API key
- Create an account if you don't have one
- Generate a new API key
If you're using a workspace other than your default workspace, you may need to authorize the models:
- Go to the DashScope Model Management Console
- Find the models you want to use (
qwen-image,qwen-image-edit,qwen-vl-max,qwen-vl-plus) - Click "Authorize" or "Subscribe" for each model
- Select your workspace/project if prompted
You can set up your API keys in one of two ways:
-
Recommended approach: Copy the
config/.env.templatefile toconfig/.envand replace the placeholders with your actual API keys:cp config/.env.template config/.env # Edit config/.env to add your API keys -
Alternative approach: Copy the
config/.env.templatefile to.envin the project root directory:cp config/.env.template .env # Edit .env to add your API keys
For international endpoint only:
DASHSCOPE_API_KEY=your_international_api_key_here
For both international and mainland China endpoints:
DASHSCOPE_API_KEY=your_international_api_key_here
DASHSCOPE_API_KEY_CHINA=your_china_api_key_here
If you only provide DASHSCOPE_API_KEY, it will be used for both regions. If you provide both, the China-specific key will be used for the mainland China endpoint.
- Add the "Qwen Text-to-Image Generator" node to your workflow
- Connect a text input with your prompt
- Configure parameters as needed (size, seed, region, etc.)
- Execute the node
- The node now outputs both the generated image and its URL
- Add the "Qwen Image-to-Image Editor" node to your workflow
- Connect one or more image inputs (image1 is required, image2 and image3 are optional)
- Provide a text instruction for editing and set region
- Execute the node
- The node now outputs both the edited image and its URL
- Add the "Qwen Vision-Language Generator" node to your workflow
- Connect an image input
- Provide a text prompt for analysis (e.g., "What is in this picture?")
- Select the Qwen-VL model (qwen-vl-max or qwen-vl-plus)
- Execute the node
- The node outputs a text description of the image
- prompt (required): The text prompt for image generation
- size: Output image resolution (1664×928, 1472×1140, 1328×1328, 1140×1472, 928×1664)
- negative_prompt: Text describing content to avoid in the image
- prompt_extend: Enable intelligent prompt rewriting for better results
- seed: Random seed for generation (0 for random)
- watermark: Add Qwen-Image watermark to output
- region: Select API endpoint (international or mainland_china)
- Outputs: IMAGE (tensor), URL (string)
- prompt (required): Text instruction for editing the image
- image1 (required): Primary input image to edit
- image2 (optional): Secondary input image for multi-image editing
- image3 (optional): Tertiary input image for multi-image editing
- negative_prompt: Text describing content to avoid in the edited image
- watermark: Add Qwen-Image watermark to output
- region: Select API endpoint (international or mainland_china)
- Outputs: IMAGE (tensor), URL (string)
- image (required): Input image to analyze
- prompt (required): Text prompt for image analysis
- model: Select Qwen-VL model (qwen-vl-max, qwen-vl-plus, qwen-vl-max-latest, qwen-vl-plus-latest)
- region: Select API endpoint (international or mainland_china)
- Outputs: STRING (text description)
Prompt: "Generate an image of a dog"
Edit an existing image:
- Original Image: an image of a dog
- Prompt: "the dog is wearing a red t-shirt and a retro glasses while eating the hamburger"
Combine multiple images for advanced editing:
-
Image1: an image of a room
-
Image2: an image showing a furniture style
-
Prompt: "Apply the furniture style from the second image to the room in the first image"
-
Result: A room with the furniture style applied
-
Image1: an image of a person
-
Image2: an image of a background scene
-
Image3: an image of an object to include
-
Prompt: "Place the person from the first image into the background scene with the object from the third image"
-
Result: A composite image with all elements combined
Analyze an image:
- Image: a photo of a dog
- Prompt: "What breed is this dog and what is it doing?"
The API key is loaded from the DASHSCOPE_API_KEY environment variable and never stored in files or code, following Alibaba Cloud security best practices.
- Enhanced Image-to-Image Editor with multi-image support (up to 3 images)
- Added image2 and image3 optional inputs to the Qwen Image-to-Image Editor node
- Updated API payload structure to support multi-image editing workflows
- Improved content array handling to accommodate multiple input images
- Enhanced debugging output to show number of image inputs
- Added new Qwen-VL generator node for image understanding and description tasks
- Supports qwen-vl-max, qwen-vl-plus, and latest model variants
- Compatible with both international and mainland China API endpoints
- Enhanced error handling with detailed error messages for common API issues
- Added stream parameter support for potential future streaming capabilities
- Enhanced API error handling with specific error messages for common issues
- Improved documentation and examples
- Initial release with core functionality
- Text-to-Image generation with qwen-image model
- Image-to-Image editing with qwen-image-edit model
- Support for international and mainland China API endpoints
- Configurable parameters for image generation and editing
This project is licensed under the MIT License - see the LICENSE file for details.