comfy-api-client

A Python client for the ComfyUI API, providing:

⚡ Full API coverage
⚡ Asynchronous execution
⚡ WebSocket support
⚡ Workflow templating
⚡ WebSocket or HTTP-based polling

Installation

Install the package using pip:

pip install comfy-api-client

Usage

Export ComfyUI workflow in API format

ComfyUI has two different workflow formats. The standard format for which contains additional meta data for the UI and a reduced API version which corresponds to the actual execution graph that is send to the backend. Exporting workflows in API format has to be enabled in the developer settings first.

API format export instructions

Afterwards, a separate button for API format export should appear below the normal "Save" button:

Create a Client

Use the create_client context manager to create a ComfyUI client. This will set up the underlying HTTP client and a WebSocket or HTTP-based state tracker to poll results from the server:

from comfy_api_client import create_client

# Protocol is omitted as the URL may be used for both HTTP and WebSocket requests
comfyui_server = "localhost:8188"

async with create_client(comfyui_server) as client:
    print(await client.get_system_stats())

Submit Workflows

To submit a workflow, read the workflow configuration file and pass it to the client:

from comfy_api_client import utils

workflow = utils.read_json("workflow.json")

async with create_client(comfyui_server) as client:
    prompt = await client.submit_workflow(workflow)

    result = await prompt.future
    image_items = result.output_images
    image = image_items[0].image

Result polling

HTTP-based polling

By default, the execution state of a prompt is tracked via a WebSocket connection. In cases where a WebSocket connection cannot be established, e.g. if the ComfyUI server is behind a proxy or firewall, HTTP-based polling can be used instead.

Simply provide start_state_tracker="http" to the create_client function:

async with create_client(comfyui_server, start_state_tracker="http") as client:
    # Submit and await results as before
    prompt = await client.submit_workflow(workflow)
    result = await prompt.future

Note, that HTTP polling relies on frequent querying of the prompt status from the ComfyUI server and will thus create more traffic while being less responsive.

Manual polling

The prompt status and results can also be queried manually. In this case, start_state_tracker=None can be passed to create_client and return_future=False to the submit_workflow() method.

import time

async with create_client(comfyui_server, start_state_tracker=None) as client:
    # Submit and await result as before
    prompt = await client.submit_workflow(workflow, return_future=False)

    # Will be None; the status needs to be checked manually
    prompt.future

    # Wait for the prompt to finish
    time.sleep(20)

    result = await client.fetch_results(prompt)

    image_items = result.output_images
    image = image_items[0].image

Tests

Run tests:

pytest tests

This will set up a local ComfyUI instance to test against.

TODOs

• Add logging support
• Improve error handling and messages
• Implement a synchronous client