ComfyUI API Client

EN | CN

A Python client library for interacting with ComfyUI API. This library provides a convenient wrapper to queue workflows, upload images/masks, and retrieve generated results (images, videos, audio, etc.) programmatically.

Features

• Workflow Management: Queue workflows easily and wait for results.
• File Management: Upload input images and masks directly to ComfyUI.
• Generic Output Handling: Automatically handles various output types (Images, Videos/GIFs, Audio).
• Execution Control: Interrupt running tasks, check queue status, and view history.
• Node Info: Retrieve definitions for ComfyUI nodes.

Installation

pip install comfyui_xy

How to Get Workflow JSON

To use this API, you need the workflow in API Format, which is different from the standard JSON saved by ComfyUI.

1. Enable Dev Mode Options:

   • In ComfyUI web interface, click the Settings (gear icon) in the menu.
   • Check the option "Enable Dev mode Options".

2. Save as API Format:

   • Once Dev Mode is enabled, you will see a new button in the menu: "Save (API Format)".
   • Click this button to save your workflow as a JSON file (e.g., workflow_api.json).
   • Use this JSON file with the ComfyUiClient.

Quick Start

import json
from comfyui_xy import ComfyUiClient

# 1. Initialize Client
client = ComfyUiClient(url="http://127.0.0.1:8188")

# 2. Load Workflow
# You should export the workflow in "API Format" from ComfyUI
with open("workflow_api.json", "r", encoding="utf-8") as f:
    workflow = json.load(f)

# 3. Modify Parameters (Optional)
# e.g., Change the seed in KSampler (Node ID "3")
import random
workflow["3"]["inputs"]["seed"] = random.randint(1, 1000000000)

# 4. Run Workflow
print("Queueing workflow...")
responses = client.process_workflow(workflow)

# 5. Handle Unified Responses
for i, response in enumerate(responses):
    if response.is_file:
        print(f"Received file: {response.filename} ({response.file_type})")
        response.save(f"output_{i}_{response.filename}")

        if response.file_type == "image":
            response.show()
    elif response.is_text:
        print(f"Received text from node {response.node_id}: {response.text}")

Detailed Usage

1. Initialization

from comfyui_xy import ComfyUiClient

# Default local server (http://127.0.0.1:8188)
client = ComfyUiClient() 

# Specify URL
client = ComfyUiClient(url="http://127.0.0.1:8188")

# Remote server with HTTPS
client = ComfyUiClient(url="https://my-comfyui-server.com:8188")

# Recommended for production scripts: explicit request timeout and strict errors
client = ComfyUiClient(timeout=30, execution_timeout=300, raise_on_error=True)

2. Uploading Files

You can upload images or masks before running a workflow. These files are saved in the input directory of ComfyUI.

# Upload an image
# Returns the filename used by ComfyUI (useful for setting node inputs)
image_name = client.upload_image("path/to/my_image.png")

# Upload a mask
mask_name = client.upload_mask("path/to/my_mask.png")

# Example: Set the uploaded image in a LoadImage node (e.g., Node ID "10")
workflow["10"]["inputs"]["image"] = image_name

3. Processing Workflows

The process_workflow method is a high-level helper that:

1. Queues the prompt.
2. Waits for execution to finish.
3. Returns unified response objects for downloaded files and discovered text outputs.

responses = client.process_workflow(workflow)

Return Value: It returns a list of ComfyResponse objects. Each response has output_type set to either "file" or "text".

If your workflow produces text instead of files, return the raw ComfyUI outputs data and extract the strings directly:

outputs = client.process_workflow(workflow, return_raw=True)
texts = client.extract_text_outputs(outputs)

print(outputs)
print(texts)

If you want explicit response objects from raw outputs, use:

file_responses = client.extract_file_responses(outputs)
text_responses = client.extract_text_responses(outputs)
responses = client.extract_outputs(outputs)

If you want failures to be explicit instead of silently returning None / {} / [], initialize the client with raise_on_error=True. Request failures raise ComfyUiError, ComfyUiHttpError, or ComfyUiTimeoutError.

4. Handling Responses (ComfyResponse)

The ComfyResponse object wraps the raw data returned by ComfyUI.

• Attributes:

  • output_type: "file" or "text".
  • data: Raw bytes of the file for file responses.
  • text: Extracted text for text responses.
  • filename: Original filename on server for file responses.
  • node_id: Source node ID in ComfyUI outputs.
  • output_name: Output field name such as images or text.
  • file_type: Type of file (e.g., 'image', 'video', 'audio', 'text').
  • image: A PIL.Image object (if the response is a valid image file).

• Methods:

  • save(path=None): Save file bytes or text content to disk.
  • show(): Opens the image in the default viewer (only for images).

5. Advanced Controls

Interrupt Execution:

client.interrupt()

Get System Status:

# Get Queue Status (Pending/Running tasks)
queue_info = client.get_queue()

# Get Full History
history = client.get_history_all()

# Get Node Information
node_info = client.get_object_info("KSampler")

Async Support

You can use AsyncComfyUiClient for asynchronous operations using aiohttp.

import asyncio
from comfyui_xy import AsyncComfyUiClient

async def main():
    async with AsyncComfyUiClient(url="http://127.0.0.1:8188") as client:
        # Most methods are awaitable
        results = await client.process_workflow(workflow)

if __name__ == "__main__":
    asyncio.run(main())

Examples

Check the examples/ directory for more complete scripts:

• basic_workflow.py: Simple Text-to-Image generation.
• image_to_image.py: Uploading an image and processing it.
• advanced_control.py: Inspecting queue and system info.

Changelog

0.3.0

• Added unified ComfyResponse outputs. process_workflow() now returns both file and text responses by default.
• Kept return_raw=True support for callers that need the original ComfyUI outputs structure.
• Added extract_file_responses(), extract_text_responses(), extract_outputs(), and extract_text_outputs() helpers for explicit post-processing.
• Added request timeout, workflow execution timeout, and strict error mode via raise_on_error=True.
• Added ComfyUiError, ComfyUiHttpError, and ComfyUiTimeoutError.
• Added sync and async test coverage for raw outputs, unified responses, HTTP errors, and timeouts.

Compatibility note:

• Older code that assumed every item returned by process_workflow() was a file should now check response.is_file before calling file-specific logic.

License

MIT