# Comfy-Pack: Making ComfyUI Workflows Shareable ![banner2](https://github.com/user-attachments/assets/14a7e469-6683-4818-9d54-5e5a8d0aa454) comfy-pack is a comprehensive toolkit for reliably packing and unpacking environments for ComfyUI workflows. - 📦 **Pack workflow environments as artifacts:** Saves the workflow environment in a `.cpack.zip` artifact with Python package versions, ComfyUI and custom node revisions, and model hashes. - ✨ **Unpack artifacts to recreate workflow environments:** Unpacks the `.cpack.zip` artifact to recreate the same environment with the exact Python package versions, ComfyUI and custom node revisions, and model weights. - 🚀 **Deploy workflows as APIs:** Deploys the workflow as a RESTful API with customizable input and output parameters. ## Motivations ComfyUI Manager is great for find missing custom nodes. But when sharing ComfyUI workflows to others(your audience or team members), you've still likely heard these responses: - "Custom Node not found" - "Cannot find the correct model file" - "Missing Python dependencies" These are fundamental challenges in workflow sharing – every component should match exactly: custom nodes, model files, and Python dependencies. Modern pacakge managers like npm and poetry introduced "lock" feature, which means record the exact version for every requirement. ComfyUI Manager isn't designed for that. We learned it from our community and developed comfy-pack to address these problems. With a single click, it captures and locks your entire workflow environment into a `.cpack.zip` file, including Python packages, custom nodes, model hashes, and required assets. Users can recreate the exact environment with one command: ```bash comfy-pack unpack workflow.cpack.zip ``` This means you can focus on your creative work while comfy-pack handles the rest. ## Usages ### Installation We recommend you use ComfyUI Manager to install comfy-pack. Simply search for `comfy-pack` and click **Install**. Restart the server and refresh your ComfyUI interface to apply changes. ![install_node](https://github.com/user-attachments/assets/dbfb730d-edff-4a52-b6c4-695e3ec70368) Alternatively, clone the project repository through `git`. ```bash cd ComfyUI/custom_nodes git clone https://github.com/bentoml/comfy-pack.git ``` To install the comfy-pack CLI, run: ```bash pip install comfy-pack ``` ### Pack a ComfyUI workflow and its environment You can package a workflow and the environment required to run the workflow into an artifact that can be unpacked elsewhere. 1. Click the **Package** button to create a `.cpack.zip` artifact. 2. (Optional) Select the models that you want to include (only model hash will be recorded, so you won't get a 100GB zip file). ![pack](https://github.com/user-attachments/assets/e08bbed2-84dc-474e-a701-6c6db16edf76) ### Unpack the ComfyUI environments Unpacking a `.cpack.zip` artifact will restore the ComfyUI environment for the workflow. During unpacking, comfy-pack will perform the following steps. 1. Prepare a Python virtual environment with the exact packages used to run the workflow. 2. Clone ComfyUI and custom nodes from the exact revisions required by the workflow. 3. Search for and download models from common registries like Hugging Face and Civitai. Unpacking workflows using the same model will not cause the model to be downloaded multiple times. Instead, model weights will be symbolically linked. To unpack: ```bash comfy-pack unpack workflow.cpack.zip ``` Huggingface gated models can be accessed by setting your `HF_TOKEN` as an environment variable before unpacking: ```bash export HF_TOKEN=hf_XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXx comfy-pack unpack workflow.cpack.zip ``` For example cpack files, check our [examples folder](examples/). ### Deploy a workflow as an API You can turn a ComfyUI workflow into an API endpoint callable using any clients through HTTP.

1. Annotate input & output

Use custom nodes provided by comfy-pack to annotate the fields to be used as input and output parameters. To add a comfy-pack node, right-click and select **Add Node** > **ComfyPack** > **output/input** > [Select a type] Input nodes: - ImageInput: Accepts `image` type input, similar to the official `LoadImage` node - StringInput: Accepts `string` type input (e.g., prompts) - IntInput: Accepts `int` type input (e.g., dimensions, seeds) - AnyInput: Accepts `combo` type and more input (e.g., custom nodes) ![input](https://github.com/user-attachments/assets/44264007-0ac8-4e23-8dc0-e60aa0ebcea2) Output nodes: - ImageOutput: Outputs `image` type, similar to the official `SaveImage` node - FileOutput: Outputs file path as `string` type and saves the file under that path ![output](https://github.com/user-attachments/assets/a4526661-8930-4575-bacc-33b6887f6271) More field types are under way.

2. Serve the workflow

Start an HTTP server at `http://127.0.0.1:3000` (default) to serve the workflow under the `/generate` path. ![serve](https://github.com/user-attachments/assets/8d4c92c5-d6d7-485e-bc71-e4fc0fe8bf35) You can call the `/generate` endpoint by specifying parameters configured through your comfy-pack nodes, such as prompt, width, height, and seed. > [!NOTE] > The name of a comfy-pack node is the parameter name used for API calls. Examples to call the endpoint: CURL ```bash curl -X 'POST' \ 'http://127.0.0.1:3000/generate' \ -H 'accept: application/octet-stream' \ -H 'Content-Type: application/json' \ -d '{ "prompt": "rocks in a bottle", "width": 512, "height": 512, "seed": 1 }' ``` BentoML client Under the hood, comfy-pack leverages [BentoML](https://github.com/bentoml/BentoML), the unified model serving framework. You can invoke the endpoint using [the BentoML Python client](https://docs.bentoml.com/en/latest/build-with-bentoml/clients.html): ```python import bentoml with bentoml.SyncHTTPClient("http://127.0.0.1:3000") as client: result = client.generate( prompt="rocks in a bottle", width=512, height=512, seed=1 ) ```

3. (Optional) Pack the workflow and environment

Pack the workflow and environment into an artifact that can be unpacked elsewhere to recreate the workflow. ```bash # Get the workflow input spec comfy-pack run workflow.cpack.zip --help # Run comfy-pack run workflow.cpack.zip --src-image image.png --video video.mp4 ```

4. (Optional) Deploy to the cloud

Deploy to [BentoCloud](https://www.bentoml.com/) with access to a variety of GPUs and blazing fast scaling. Follow [the instructions here](https://docs.bentoml.com/en/latest/scale-with-bentocloud/manage-api-tokens.html) to get your BentoCloud access token. If you don’t have a BentoCloud account, you can [sign up for free](https://bentoml.com/). ![image](https://github.com/user-attachments/assets/1ffa31fc-1f50-4ea7-a47e-7dae3b874273)

## Security Guidelines A cpack file only contains the metadata of the workflow environment, such as Python package versions, ComfyUI and custom node revisions, and model hashes. It does not contain any sensitive information like API keys, passwords, or user data. However, unpacking a cpack file will install custom nodes and Python dependencies. It is recommended to unpack cpack files from trusted sources. comfy-pack has a strict mode for unpacking. You can enable it by setting the `CPACK_STRICT_MODE` environment variable to `true`. It will sacrifice some flexibility and compatibility for security. For now, comfy-pack will: * Use more strict index strategy in Python package installation More security features are under way. ## Roadmap This project is under active development. Currently we are working on: - Enhanced user experience - Docker support - Local `.cpack` file management with version control - Enhanced service capabilities ## Community comfy-pack is actively maintained by the BentoML team. Feel free to reach out 👉 [Join our Slack community!](https://l.bentoml.com/join-slack) ## Contributing As an open-source project, we welcome contributions of all kinds, such as new features, bug fixes, and documentation. Here are some of the ways to contribute: - Repost a bug by creating a [GitHub issue](https://github.com/bentoml/comfy-pack/issues). - Submit a [pull request](https://github.com/bentoml/comfy-pack/pulls) or help review other developers’ pull requests.

Name	Type	Description
IMAGE	IMAGE	—
MASK	MASK	—

Image Input

Inputs (1)

Outputs (2)