comfy-proxy

A Python client library for interacting with ComfyUI servers and managing image generation workflows.

Features

• Single and multi-instance ComfyUI server management
• Async workflow execution
• Flexible address configuration (single, multiple, port ranges)
• Built-in workflow templates
• Customizable image sizes and parameters
• Websocket-based image generation and retrieval

Installation

ComfyProxy requires Python 3.11 or later. Install using pip:

pip install comfy-proxy

Quick Start

import asyncio
from comfy_proxy.comfy import SingleComfy
from comfy_proxy.workflows.flux import FluxModel, FluxWorkflowParams, FluxWorkflow
from comfy_proxy.workflow import Sizes

async def generate_image():
    # Initialize a ComfyUI instance
    comfy = Comfy("127.0.0.1:7821")
    
    # Configure model and parameters
    model = FluxModel()
    params = FluxWorkflowParams(
        prompt="a beautiful sunset over mountains",
        model=model,
        size=Sizes.LANDSCAPE_16_9,
        seed=42,
        batch_size=1
    )
    
    # Create and execute workflow
    workflow = FluxWorkflow(params)
    async for image_data in comfy.generate(workflow):
        with open("generated_image.png", "wb") as f:
            f.write(image_data)
        break  # Save first image only

# Run the example
asyncio.run(generate_image())

Multiple ComfyUI Instances

ComfyProxy can distribute work across multiple ComfyUI servers:

from comfy_proxy.comfy import Comfy

# Initialize with multiple instances
comfy = Comfy("127.0.0.1:8188-8192")  # Port range
# or
comfy = Comfy(["127.0.0.1:8188", "127.0.0.1:8189"])  # List
# or
comfy = Comfy("127.0.0.1:8188,127.0.0.1:8189")  # Comma-separated

Development

Requirements:

• Python 3.11+
• Poetry for dependency management

Setup:

# Clone repository
git clone https://github.com/yourusername/comfy-proxy.git
cd comfy-proxy

# Install dependencies
poetry install

# Run tests
poetry run pytest

Built-In Workflows

• Flux t2i
• Wan 2.2 i2v (first frame, first+last frame, lightning)
• Qwen Image (regular, lightning)
• Qwen Image Edit 2509 (regular, lightning)

These correspond roughly to the default ComfyUI workflows and should be highly compatible.

Custom Workflows

ComfyProxy comes with some built-in workflows (like FluxWorkflow), but you can easily create your own:

from comfy_proxy.workflow import ComfyWorkflow, ComfyNode

class MyCustomWorkflow(ComfyWorkflow):
    def __init__(self, params):
        super().__init__()
        self.params = params
        self._build_workflow()
        
    def _build_workflow(self):
        # Add nodes to create your workflow
        checkpoint_node = self.add_node(
            "CheckpointLoaderSimple",
            {"ckpt_name": "sd_xl_base_1.0.safetensors"}
        )
        # ... add more nodes and connect them

We welcome contributions of basic workflows for popular models:

• Stable Diffusion 1.5
• SDXL
• SD Turbo
• SD 3.0
• etc.

For more complex use cases, you can implement the ComfyWorkflow class directly in your own codebase to create custom workflows with additional functionality.

License

MIT License - see LICENSE file for details.