A vanilla, up-to-date fork of ComfyUI.

• Run with comfyui in your command line.
• Installable via pip: pip install git+https://github.com/hiddenswitch/ComfyUI.git.
• Distributed with support for multiple GPUs, multiple backends and frontends, including in containers, using RabbitMQ.
• Installable custom nodes via pip.
• New configuration options for directories, models and metrics.
• API support, using the vanilla ComfyUI API and new API endpoints.
• Embed ComfyUI as a library inside your Python application. No server or frontend needed.
• Containers for running on Linux, Windows and Kubernetes with CUDA acceleration.
• Automated tests for new features.
• Automatic model downloading for well-known models.

• Workflows
• Installing
• Configuration
• Installing Custom Nodes
• Authoring Custom Nodes
• API
• Distributed

You must have Python 3.10, 3.11 or 3.12 installed. On Windows, download the latest Python from their website. You can also directly download 3.11.4 here.

On macOS, install exactly Python 3.11 using brew, which you can download from https://brew.sh, using this command: brew install [email protected]. Do not use 3.9 or older, and do not use 3.12 or newer. Its compatibility with Stable Diffusion in both directions is broken.

1. Create a virtual environment:

   python -m virtualenv venv

2. Activate it on Windows (PowerShell):

Set-ExecutionPolicy Unrestricted -Scope Process
& .\venv\Scripts\activate.ps1

Linux and macOS

source ./venv/bin/activate

3. Then, run the following command to install comfyui into your current environment. This will correctly select the version of pytorch that matches the GPU on your machine (NVIDIA or CPU on Windows, NVIDIA AMD or CPU on Linux):

   pip install git+https://github.com/hiddenswitch/ComfyUI.git

   Advanced: If you are running in Google Collab or another environment which has already installed torch for you, disable build isolation, and the package will recognize your currently installed torch.

   # You will need wheel, which isn't included in Python 3.11 or later
   pip install wheel
   pip install --no-build-isolation git+https://github.com/hiddenswitch/ComfyUI.git

4. Create the directories you can fill with checkpoints:

   comfyui --create-directories

   Your current working directory is wherever you started running comfyui. You don't need to clone this repository, observe it is omitted from the instructions. You can cd into a different directory containing models/, or if the models are located somehwere else, like C:/some directory/models, do:

   comfyui --cwd="C:/some directory/"

   You can see all the command line options with hints using comfyui --help.

5. To run the web server:

   comfyui

   When you run workflows that use well-known models, this will download them automatically.

   To make it accessible over the network:

   comfyui --listen

These models will be automatically downloaded when you queue prompts with workflows that use them:

• Stable Diffusion XL (SDXL) Base: stabilityai/stable-diffusion-xl-base-1.0
• Stable Diffusion XL (SDXL) Refiner: stabilityai/stable-diffusion-xl-refiner-1.0
• SDXL Turbo: stabilityai/sdxl-turbo (both FP16 and standard versions)
• Stable Cascade: The stages from stabilityai/stable-cascade (Stage A, Stage B, and Stage C)
• Stable Diffusion 1.5: From runwayml/stable-diffusion-v1-5 (including pruned, EMA-only, and pruned versions)
• Stable Diffusion 2.1: stabilityai/stable-diffusion-2-1 (v2-1 768 EMA pruned version)
• Waifu Diffusion 1.5 Beta3: waifu-diffusion/wd-1-5-beta3 (Illusion FP16 version)
• NeverEnding Dream: jomcs/NeverEnding_Dream-Feb19-2023 (CarDos Anime version)
• Anything-V3.0: The anime model ckpt/anything-v3.0
• Stable Diffusion 2.1 Unclip: stabilityai/stable-diffusion-2-1-unclip (both high and low versions)
• Stable Zero123: stabilityai/stable-zero123
• Annotators (RealESRGAN Upscaler): lllyasviel/Annotators (RealESRGAN x4plus)
• GLIGEN Pruned Safetensors: comfyanonymous/GLIGEN_pruned_safetensors (both standard and FP16 versions)
• CLIP Vision G: comfyanonymous/clip_vision_g

1. Clone this repo:

   git clone https://github.com/comfyanonymous/ComfyUI.git
   cd ComfyUI

2. Create a virtual environment:

   1. Create an environment:

      python -m virtualenv venv

   2. Activate it:

      Windows (PowerShell):

      Set-ExecutionPolicy Unrestricted -Scope Process
      & .\venv\Scripts\activate.ps1

      Linux and macOS

      source ./venv/bin/activate

3. Then, run the following command to install comfyui into your current environment. This will correctly select the version of pytorch that matches the GPU on your machine (NVIDIA or CPU on Windows, NVIDIA AMD or CPU on Linux):

   pip install -e ".[dev]"

4. To run the web server:

   comfyui

   To run tests:

   pytest tests/inference
   (cd tests-ui && npm ci && npm run test:generate && npm test)

   You can use comfyui as an API. Visit the OpenAPI specification. This file can be used to generate typed clients for your preferred language.

5. To create the standalone binary:

   python -m PyInstaller --onefile --noupx -n ComfyUI --add-data="comfy/;comfy/" --paths $(pwd) --paths comfy/cmd main.py

Because pip installs the package as editable with pip install -e ., any changes you make to the repository will affect the next launch of comfy. In IDEA based editors like PyCharm and IntelliJ, the Relodium plugin supports modifying your custom nodes or similar code while the server is running.

Custom Nodes can be added to ComfyUI by copying and pasting Python files into your ./custom_nodes directory.

There are two kinds of custom nodes: vanilla custom nodes, which generally expect to be dropped into the custom_nodes directory and managed by a tool called the ComfyUI Extension manager ("vanilla" custom nodes) and this repository's opinionated, installable custom nodes ("installable").

Clone the repository containing the custom nodes into custom_nodes/ in your working directory. Currently, this is not known to be compatible with ComfyUI Node Manager.

Run pip install git+https://github.com/owner/repository, replacing the git repository with the installable custom nodes URL. This is just the GitHub URL.

Create a requirements.txt:

comfyui

Observe comfyui is now a requirement for using your custom nodes. This will ensure you will be able to access comfyui as a library. For example, your code will now be able to import the folder paths using from comfyui.cmd import folder_paths. Because you will be using my fork, use this:

comfyui @ git+https://github.com/hiddenswitch/ComfyUI.git

Additionally, create a pyproject.toml:

[build-system]
requires = ["setuptools", "wheel", "pip"]
build-backend = "setuptools.build_meta"

This ensures you will be compatible with later versions of Python.

Finally, move your nodes to a directory with an empty __init__.py, i.e., a package. You should have a file structure like this:

# the root of your git repository
/.git
/pyproject.toml
/requirements.txt
/mypackage_custom_nodes/__init__.py
/mypackage_custom_nodes/some_nodes.py

Finally, create a setup.py at the root of your custom nodes package / repository. Here is an example:

setup.py

from setuptools import setup, find_packages
import os.path

setup(
    name="mypackage",
    version="0.0.1",
    packages=find_packages(),
    install_requires=open(os.path.join(os.path.dirname(__file__), "requirements.txt")).readlines(),
    author='',
    author_email='',
    description='',
    entry_points={
        'comfyui.custom_nodes': [
            'mypackage = mypackage_custom_nodes',
        ],
    },
)

All .py files located in the package specified by the entrypoint with your package's name will be scanned for node class mappings declared like this:

some_nodes.py:

from comfy.nodes.package_typing import CustomNode

class Binary_Preprocessor(CustomNode):
  ...

NODE_CLASS_MAPPINGS = {
    "BinaryPreprocessor": Binary_Preprocessor
}
NODE_DISPLAY_NAME_MAPPINGS = {
    "BinaryPreprocessor": "Binary Lines"
}

These packages will be scanned recursively.

Extending the comfy.nodes.package_typing.CustomNode provides type hints for authoring nodes.

Declare an entry point for configuration hooks in your setup.py that defines a function that takes and returns an configargparser.ArgParser object:

setup.py

setup(
  name="mypackage",
  ...
  entry_points={
    'comfyui.custom_nodes': [
      'mypackage = mypackage_custom_nodes',
    ],
    'comfyui.custom_config': [
      'mypackage = mypackage_custom_config:add_configuration',
    ]
  },
)

mypackage_custom_config.py:

import configargparse


def add_configuration(parser: configargparse.ArgParser) -> configargparse.ArgParser:
  parser.add_argument("--openai-api-key",
                      required=False,
                      type=str,
                      help="Configures the OpenAI API Key for the OpenAI nodes", env_var="OPENAI_API_KEY")
  return parser

You can now see your configuration option at the bottom of the --help command along with hints for how to use it:

$ comfyui --help
usage: comfyui.exe [-h] [-c CONFIG_FILE] [--write-out-config-file CONFIG_OUTPUT_PATH] [-w CWD] [-H [IP]] [--port PORT]
                   [--enable-cors-header [ORIGIN]] [--max-upload-size MAX_UPLOAD_SIZE] [--extra-model-paths-config PATH [PATH ...]]
...
                   [--openai-api-key OPENAI_API_KEY]

options:
  -h, --help            show this help message and exit
  -c CONFIG_FILE, --config CONFIG_FILE
                        config file path
  --write-out-config-file CONFIG_OUTPUT_PATH
                        takes the current command line args and writes them out to a config file at the given path, then exits
  -w CWD, --cwd CWD     Specify the working directory. If not set, this is the current working directory. models/, input/, output/ and other
                        directories will be located here by default. [env var: COMFYUI_CWD]
  -H [IP], --listen [IP]
                        Specify the IP address to listen on (default: 127.0.0.1). If --listen is provided without an argument, it defaults to
                        0.0.0.0. (listens on all) [env var: COMFYUI_LISTEN]
  --port PORT           Set the listen port. [env var: COMFYUI_PORT]
...
  --distributed-queue-name DISTRIBUTED_QUEUE_NAME
                        This name will be used by the frontends and workers to exchange prompt requests and replies. Progress updates will be
                        prefixed by the queue name, followed by a '.', then the user ID [env var: COMFYUI_DISTRIBUTED_QUEUE_NAME]
  --external-address EXTERNAL_ADDRESS
                        Specifies a base URL for external addresses reported by the API, such as for image paths. [env var:
                        COMFYUI_EXTERNAL_ADDRESS]
  --openai-api-key OPENAI_API_KEY
                        Configures the OpenAI API Key for the OpenAI nodes [env var: OPENAI_API_KEY]

You can now start comfyui with:

comfyui --openai-api-key=abcdefg12345

or set the environment variable you specified:

export OPENAI_API_KEY=abcdefg12345
comfyui

or add it to your config file:

config.yaml:

openapi-api-key: abcdefg12345

comfyui --config config.yaml

Since comfyui looks for a config.yaml in your current working directory by default, you can omit the argument if config.yaml is located in your current working directory:

comfyui

Your entry point for adding configuration options should not import your nodes. This gives you the opportunity to use the configuration you added in your nodes; otherwise, if you imported your nodes in your configuration entry point, the nodes will potentially be initialized without any configuration.

Access your configuration from cli_args:

from comfy.cli_args import args
from comfy.cli_args_types import Configuration
from typing import Optional


# Add type hints when accessing args
class CustomConfiguration(Configuration):
  def __init__(self):
    super().__init__()
    self.openai_api_key: Optional[str] = None


args: CustomConfiguration


class OpenAINode(CustomNode):
  ...

  def execute(self):
    openai_api_key = args.open_api_key

I see a message like RuntimeError: '"upsample_bilinear2d_channels_last" not implemented for 'Half''

You must use Python 3.11 on macOS devices, and update to at least Ventura.

I see a message like Error while deserializing header: HeaderTooLarge

Download your model file again.

Only parts of the graph that have an output with all the correct inputs will be executed.

Only parts of the graph that change from each execution to the next will be executed, if you submit the same graph twice only the first will be executed. If you change the last part of the graph only the part you changed and the part that depends on it will be executed.

Dragging a generated png on the webpage or loading one will give you the full workflow including seeds that were used to create it.

You can use () to change emphasis of a word or phrase like: (good code:1.2) or (bad code:0.8). The default emphasis for () is 1.1. To use () characters in your actual prompt escape them like \( or \).

You can use {day|night}, for wildcard/dynamic prompts. With this syntax "{wild|card|test}" will be randomly replaced by either "wild", "card" or "test" by the frontend every time you queue the prompt. To use {} characters in your actual prompt escape them like: \{ or \}.

Dynamic prompts also support C-style comments, like // comment or /* comment */.

To use a textual inversion concepts/embeddings in a text prompt put them in the models/embeddings directory and use them in the CLIPTextEncode node like this (you can omit the .pt extension):

embedding:embedding_filename.pt

Make sure you use the regular loaders/Load Checkpoint node to load checkpoints. It will auto pick the right settings depending on your GPU.

You can set this command line setting to disable the upcasting to fp32 in some cross attention operations which will increase your speed. Note that this will very likely give you black images on SD2.x models. If you use xformers or pytorch attention this option does not do anything.

--dont-upcast-attention

Use --preview-method auto to enable previews.

The default installation includes a fast latent preview method that's low-resolution. To enable higher-quality previews with TAESD, download the taesd_decoder.pth (for SD1.x and SD2.x) and taesdxl_decoder.pth (for SDXL) models and place them in the models/vae_approx folder. Once they're installed, restart ComfyUI to enable high-quality previews.

Ctrl can also be replaced with Cmd instead for macOS users

usage: comfyui.exe [-h] [-c CONFIG_FILE]
                   [--write-out-config-file CONFIG_OUTPUT_PATH] [-w CWD]
                   [-H [IP]] [--port PORT] [--enable-cors-header [ORIGIN]]
                   [--max-upload-size MAX_UPLOAD_SIZE]
                   [--extra-model-paths-config PATH [PATH ...]]
                   [--output-directory OUTPUT_DIRECTORY]
                   [--temp-directory TEMP_DIRECTORY]
                   [--input-directory INPUT_DIRECTORY] [--auto-launch]
                   [--disable-auto-launch] [--cuda-device DEVICE_ID]
                   [--cuda-malloc | --disable-cuda-malloc]
                   [--dont-upcast-attention] [--force-fp32 | --force-fp16]
                   [--bf16-unet | --fp16-unet | --fp8_e4m3fn-unet | --fp8_e5m2-unet]
                   [--fp16-vae | --fp32-vae | --bf16-vae] [--cpu-vae]
                   [--fp8_e4m3fn-text-enc | --fp8_e5m2-text-enc | --fp16-text-enc | --fp32-text-enc]
                   [--directml [DIRECTML_DEVICE]] [--disable-ipex-optimize]
                   [--preview-method [none,auto,latent2rgb,taesd]]
                   [--use-split-cross-attention | --use-quad-cross-attention | --use-pytorch-cross-attention]
                   [--disable-xformers]
                   [--gpu-only | --highvram | --normalvram | --lowvram | --novram | --cpu]
                   [--disable-smart-memory] [--deterministic]
                   [--dont-print-server] [--quick-test-for-ci]
                   [--windows-standalone-build] [--disable-metadata]
                   [--multi-user] [--create-directories]
                   [--plausible-analytics-base-url PLAUSIBLE_ANALYTICS_BASE_URL]
                   [--plausible-analytics-domain PLAUSIBLE_ANALYTICS_DOMAIN]
                   [--analytics-use-identity-provider]
                   [--distributed-queue-connection-uri DISTRIBUTED_QUEUE_CONNECTION_URI]
                   [--distributed-queue-worker] [--distributed-queue-frontend]
                   [--distributed-queue-name DISTRIBUTED_QUEUE_NAME]

options:
  -h, --help            show this help message and exit
  -c CONFIG_FILE, --config CONFIG_FILE
                        config file path
  --write-out-config-file CONFIG_OUTPUT_PATH
                        takes the current command line args and writes them
                        out to a config file at the given path, then exits
  -w CWD, --cwd CWD     Specify the working directory. If not set, this is the
                        current working directory. models/, input/, output/
                        and other directories will be located here by default.
                        [env var: COMFYUI_CWD]
  -H [IP], --listen [IP]
                        Specify the IP address to listen on (default:
                        127.0.0.1). If --listen is provided without an
                        argument, it defaults to 0.0.0.0. (listens on all)
                        [env var: COMFYUI_LISTEN]
  --port PORT           Set the listen port. [env var: COMFYUI_PORT]
  --enable-cors-header [ORIGIN]
                        Enable CORS (Cross-Origin Resource Sharing) with
                        optional origin or allow all with default '*'. [env
                        var: COMFYUI_ENABLE_CORS_HEADER]
  --max-upload-size MAX_UPLOAD_SIZE
                        Set the maximum upload size in MB. [env var:
                        COMFYUI_MAX_UPLOAD_SIZE]
  --extra-model-paths-config PATH [PATH ...]
                        Load one or more extra_model_paths.yaml files. [env
                        var: COMFYUI_EXTRA_MODEL_PATHS_CONFIG]
  --output-directory OUTPUT_DIRECTORY
                        Set the ComfyUI output directory. [env var:
                        COMFYUI_OUTPUT_DIRECTORY]
  --temp-directory TEMP_DIRECTORY
                        Set the ComfyUI temp directory (default is in the
                        ComfyUI directory). [env var: COMFYUI_TEMP_DIRECTORY]
  --input-directory INPUT_DIRECTORY
                        Set the ComfyUI input directory. [env var:
                        COMFYUI_INPUT_DIRECTORY]
  --auto-launch         Automatically launch ComfyUI in the default browser.
                        [env var: COMFYUI_AUTO_LAUNCH]
  --disable-auto-launch
                        Disable auto launching the browser. [env var:
                        COMFYUI_DISABLE_AUTO_LAUNCH]
  --cuda-device DEVICE_ID
                        Set the id of the cuda device this instance will use.
                        [env var: COMFYUI_CUDA_DEVICE]
  --cuda-malloc         Enable cudaMallocAsync (enabled by default for torch
                        2.0 and up). [env var: COMFYUI_CUDA_MALLOC]
  --disable-cuda-malloc
                        Disable cudaMallocAsync. [env var:
                        COMFYUI_DISABLE_CUDA_MALLOC]
  --dont-upcast-attention
                        Disable upcasting of attention. Can boost speed but
                        increase the chances of black images. [env var:
                        COMFYUI_DONT_UPCAST_ATTENTION]
  --force-fp32          Force fp32 (If this makes your GPU work better please
                        report it). [env var: COMFYUI_FORCE_FP32]
  --force-fp16          Force fp16. [env var: COMFYUI_FORCE_FP16]
  --bf16-unet           Run the UNET in bf16. This should only be used for
                        testing stuff. [env var: COMFYUI_BF16_UNET]
  --fp16-unet           Store unet weights in fp16. [env var:
                        COMFYUI_FP16_UNET]
  --fp8_e4m3fn-unet     Store unet weights in fp8_e4m3fn. [env var:
                        COMFYUI_FP8_E4M3FN_UNET]
  --fp8_e5m2-unet       Store unet weights in fp8_e5m2. [env var:
                        COMFYUI_FP8_E5M2_UNET]
  --fp16-vae            Run the VAE in fp16, might cause black images. [env
                        var: COMFYUI_FP16_VAE]
  --fp32-vae            Run the VAE in full precision fp32. [env var:
                        COMFYUI_FP32_VAE]
  --bf16-vae            Run the VAE in bf16. [env var: COMFYUI_BF16_VAE]
  --cpu-vae             Run the VAE on the CPU. [env var: COMFYUI_CPU_VAE]
  --fp8_e4m3fn-text-enc
                        Store text encoder weights in fp8 (e4m3fn variant).
                        [env var: COMFYUI_FP8_E4M3FN_TEXT_ENC]
  --fp8_e5m2-text-enc   Store text encoder weights in fp8 (e5m2 variant). [env
                        var: COMFYUI_FP8_E5M2_TEXT_ENC]
  --fp16-text-enc       Store text encoder weights in fp16. [env var:
                        COMFYUI_FP16_TEXT_ENC]
  --fp32-text-enc       Store text encoder weights in fp32. [env var:
                        COMFYUI_FP32_TEXT_ENC]
  --directml [DIRECTML_DEVICE]
                        Use torch-directml. [env var: COMFYUI_DIRECTML]
  --disable-ipex-optimize
                        Disables ipex.optimize when loading models with Intel
                        GPUs. [env var: COMFYUI_DISABLE_IPEX_OPTIMIZE]
  --preview-method [none,auto,latent2rgb,taesd]
                        Default preview method for sampler nodes. [env var:
                        COMFYUI_PREVIEW_METHOD]
  --use-split-cross-attention
                        Use the split cross attention optimization. Ignored
                        when xformers is used. [env var:
                        COMFYUI_USE_SPLIT_CROSS_ATTENTION]
  --use-quad-cross-attention
                        Use the sub-quadratic cross attention optimization .
                        Ignored when xformers is used. [env var:
                        COMFYUI_USE_QUAD_CROSS_ATTENTION]
  --use-pytorch-cross-attention
                        Use the new pytorch 2.0 cross attention function. [env
                        var: COMFYUI_USE_PYTORCH_CROSS_ATTENTION]
  --disable-xformers    Disable xformers. [env var: COMFYUI_DISABLE_XFORMERS]
  --gpu-only            Store and run everything (text encoders/CLIP models,
                        etc... on the GPU). [env var: COMFYUI_GPU_ONLY]
  --highvram            By default models will be unloaded to CPU memory after
                        being used. This option keeps them in GPU memory. [env
                        var: COMFYUI_HIGHVRAM]
  --normalvram          Used to force normal vram use if lowvram gets
                        automatically enabled. [env var: COMFYUI_NORMALVRAM]
  --lowvram             Split the unet in parts to use less vram. [env var:
                        COMFYUI_LOWVRAM]
  --novram              When lowvram isn't enough. [env var: COMFYUI_NOVRAM]
  --cpu                 To use the CPU for everything (slow). [env var:
                        COMFYUI_CPU]
  --disable-smart-memory
                        Force ComfyUI to agressively offload to regular ram
                        instead of keeping models in vram when it can. [env
                        var: COMFYUI_DISABLE_SMART_MEMORY]
  --deterministic       Make pytorch use slower deterministic algorithms when
                        it can. Note that this might not make images
                        deterministic in all cases. [env var:
                        COMFYUI_DETERMINISTIC]
  --dont-print-server   Don't print server output. [env var:
                        COMFYUI_DONT_PRINT_SERVER]
  --quick-test-for-ci   Quick test for CI. [env var:
                        COMFYUI_QUICK_TEST_FOR_CI]
  --windows-standalone-build
                        Windows standalone build: Enable convenient things
                        that most people using the standalone windows build
                        will probably enjoy (like auto opening the page on
                        startup). [env var: COMFYUI_WINDOWS_STANDALONE_BUILD]
  --disable-metadata    Disable saving prompt metadata in files. [env var:
                        COMFYUI_DISABLE_METADATA]
  --multi-user          Enables per-user storage. [env var:
                        COMFYUI_MULTI_USER]
  --create-directories  Creates the default models/, input/, output/ and temp/
                        directories, then exits. [env var:
                        COMFYUI_CREATE_DIRECTORIES]
  --plausible-analytics-base-url PLAUSIBLE_ANALYTICS_BASE_URL
                        Enables server-side analytics events sent to the
                        provided URL. [env var:
                        COMFYUI_PLAUSIBLE_ANALYTICS_BASE_URL]
  --plausible-analytics-domain PLAUSIBLE_ANALYTICS_DOMAIN
                        Specifies the domain name for analytics events. [env
                        var: COMFYUI_PLAUSIBLE_ANALYTICS_DOMAIN]
  --analytics-use-identity-provider
                        Uses platform identifiers for unique visitor
                        analytics. [env var:
                        COMFYUI_ANALYTICS_USE_IDENTITY_PROVIDER]
  --distributed-queue-connection-uri DISTRIBUTED_QUEUE_CONNECTION_URI
                        EXAMPLE: "amqp://guest:[email protected]" - Servers and
                        clients will connect to this AMPQ URL to form a
                        distributed queue and exchange prompt execution
                        requests and progress updates. [env var:
                        COMFYUI_DISTRIBUTED_QUEUE_CONNECTION_URI]
  --distributed-queue-worker
                        Workers will pull requests off the AMQP URL. [env var:
                        COMFYUI_DISTRIBUTED_QUEUE_WORKER]
  --distributed-queue-frontend
                        Frontends will start the web UI and connect to the
                        provided AMQP URL to submit prompts. [env var:
                        COMFYUI_DISTRIBUTED_QUEUE_FRONTEND]
  --distributed-queue-name DISTRIBUTED_QUEUE_NAME
                        This name will be used by the frontends and workers to
                        exchange prompt requests and replies. Progress updates
                        will be prefixed by the queue name, followed by a '.',
                        then the user ID [env var:
                        COMFYUI_DISTRIBUTED_QUEUE_NAME]

Args that start with '--' can also be set in a config file (config.yaml or
config.json or specified via -c). Config file syntax allows: key=value,
flag=true, stuff=[a,b,c] (for details, see syntax at https://goo.gl/R74nmi).
In general, command-line values override environment variables which override
config file values which override defaults.

There are multiple ways to use this ComfyUI package to run workflows programmatically:

Start ComfyUI by creating an ordinary Python object. This does not create a web server. It runs ComfyUI as a library, like any other package you are familiar with:

from comfy.client.embedded_comfy_client import EmbeddedComfyClient
async with EmbeddedComfyClient() as client:
  # This will run your prompt
  outputs = await client.queue_prompt(prompt)
  # At this point, your prompt is finished and all the outputs, like saving images, have been completed.
  # Now the outputs will contain the same thing that the Web UI expresses: a file path for each output.
  # Let's find the node ID of the first SaveImage node. This will work when you change your workflow JSON from
  # the example above.
  save_image_node_id = next(key for key in prompt if prompt[key].class_type == "SaveImage")
  # Now let's print the absolute path to the image.
  print(outputs[save_image_node_id]["images"][0]["abs_path"])
# At this point, all the models have been unloaded from VRAM, and everything has been cleaned up.

See script_examples/basic_api_example.py for a complete example.

Start ComfyUI as a remote server, then access it via an API. This requires you to start ComfyUI somewhere. Then access it via a standardized API.

from comfy.client.aio_client import AsyncRemoteComfyClient
client = AsyncRemoteComfyClient(server_address="http://localhost:8188")
# Now let's get the bytes of the PNG image saved by the SaveImage node:
png_image_bytes = await client.queue_prompt(prompt)
# You can save these bytes wherever you need!
with open("image.png", "rb") as f:
    f.write(png_image_bytes)

See script_examples/remote_api_example.py for a complete example.

Use a typed, generated API client for your programming language and access ComfyUI server remotely as an API. You can generate the client from comfy/api/openapi.yaml.

Submit jobs directly to a distributed work queue. This package supports AMQP message queues like RabbitMQ. You can submit workflows to the queue, including from the web using RabbitMQ's STOMP support, and receive realtime progress updates from multiple workers. Continue to the next section for more details.

This package supports multi-processing across machines using RabbitMQ. This means you can launch multiple ComfyUI backend workers and queue prompts against them from multiple frontends.

ComfyUI has two roles: worker and frontend. An unlimited number of workers can consume and execute workflows (prompts) in parallel; and an unlimited number of frontends can submit jobs. All of the frontends' API calls will operate transparently against your collection of workers, including progress notifications from the websocket.

To share work among multiple workers and frontends, ComfyUI uses RabbitMQ or any AMQP-compatible message queue like SQS or Kafka.

On a machine in your local network, install Docker and run RabbitMQ:

docker run -it --rm --name rabbitmq -p 5672:5672 rabbitmq:latest

Find the machine's main LAN IP address:

Windows (PowerShell):

Get-NetIPConfiguration | Where-Object { $_.InterfaceAlias -like '*Ethernet*' -and $_.IPv4DefaultGateway -ne $null } | ForEach-Object { $_.IPv4Address.IPAddress }

Linux

ip -4 addr show $(ip route show default | awk '/default/ {print $5}') | grep -oP 'inet \K[\d.]+'

macOS

ifconfig $(route get default | grep interface | awk '{print $2}') | awk '/inet / {print $2; exit}'

On my machine, this prints 10.1.0.100, which is a local LAN IP that other hosts on my network can reach.

On this machine, you can also set up a file share for models, outputs and inputs.

Once you have installed this Python package following the installation steps, you can start a worker using:

Starting a Worker:

# you must replace the IP address with the one you printed above
comfyui-worker --distributed-queue-connection-uri="amqp://guest:[email protected]"

All the normal command line arguments are supported. This means you can use --cwd to point to a file share containing the models/ directory:

comfyui-worker --cwd //10.1.0.100/shared/workspace --distributed-queue-connection-uri="amqp://guest:[email protected]"

Starting a Frontend:

comfyui --listen --distributed-queue-connection-uri="amqp://guest:[email protected]" --distributed-queue-frontend

However, the frontend will not be able to find the output images or models to show the client by default. You must specify a place where the frontend can find the same outputs and models that are available to the backends:

comfyui --cwd //10.1.0.100/shared/workspace --listen --distributed-queue-connection-uri="amqp://guest:[email protected]" --distributed-queue-frontend

You can carefully mount network directories into outputs/ and inputs/ such that they are shared among workers and frontends; you can store the models/ on each machine, or serve them over a file share too.

The frontend expects to find the referenced output images in its --output-directory or in the default outputs/ under --cwd (aka the "workspace").

This means that workers and frontends do not have to have the same argument to --cwd. The paths that are passed to the frontend, such as the inputs/ and outputs/ directories, must have the same contents as the paths passed as those directories to the workers.

Since reading models like large checkpoints over the network can be slow, you can use --extra-model-paths-config to specify additional model paths. Or, you can use --cwd some/path, where some/path is a local directory, and, and mount some/path/outputs to a network directory.

Known models listed in model_downloader.py are downloaded using huggingface_hub with the default cache_dir. This means you can mount a read-write-many volume, like an SMB share, into the default cache directory. Read more about this here.

Build the Dockerfile:

docker build . -t hiddenswitch/comfyui

To run:

docker run -it -v ./output:/workspace/output -v ./models:/workspace/models --gpus=all --ipc=host --ulimit memlock=-1 --ulimit stack=67108864 --rm hiddenswitch/comfyui

Chat on Matrix: #comfyui_space:matrix.org, an alternative to Discord.