NozyIO

Python AI / ML pipeline visualization tool. Automatically discover your functions as pipeline nodes. Fastest way to demo your python program for customers to tweak and expand. Perfect for AI/ML engineers and designers to collaborate, turning ML components into UI nodes for easy tweaking.

It can also be used as a visual scripting tool for python.

demo: https://youtu.be/L_6kY-fhIcU

• Automatically discover python functions as nodes with parameters typing
• Visually pick input files, tune parameters, and preview any image input/output from the GUI
• Save pipeline graph as json and switch between pipelines quickly
• Export the pipeline graph as Python code

Install

In your python project root:

pip install nozyio

To start the nozyio UI:

nozyio

Usage

Super easy node define (just write a function with typing) For example:

from PIL.Image import Image

def resize_image(
    image: Image,
    width: int = 512,
    height: int = 768,
    method: Literal["stretch", "fit", "crop"],
    interpolation: str
) -> Image:
    # ...some code here...
    return image.resize((width, height), interp_method)

resize_image.NOZY_NODE_DEF = {
    "node_title": "Resize Image",
}

👇This function will be rendered as below. You can see that all args default values are populated as the input box defaults

👇Export workflow to code, preview any image input/output

👇Automatically scan your python functions as nodes with params typing

👇Double-click canvas to search any functions in your project and add nodes

Node input types

Nozyio will automatically scan your python functions and convert them to nodes. You can define the input, output types by adding type annotations to the function parameters and return type. Params with no type annotation will become "any" type.

Input widgets - file picker

You can also add custom UI widgets to the input parameters by adding a widget field to the input definition. In below example, we use server_file_picker widget to let user select an image file from the files on the server:

from PIL import Image.Image
def load_image(image_path: str) -> Image.Image:
    return Image.open(image_path)
load_image.NOZY_NODE_DEF = {
    "node_title": "Load Image",
    "description": "Load image from path",
    "inputs": {
        "image_path": {
            "type": "filepath",
            "widget": {
                'type': 'server_file_picker',
                'options': {
                    'extensions': ['.png', '.jpg', '.jpeg', '.bmp', '.tiff', '.webp']
                }
            },
            "hide_handle": True,
            "description": "Path to image"
        }
    },
    "outputs": [{"name": "image", "type": "Image", "description": "Loaded image"}]
}

Future Plans

• Visualize your python code to graph flow
• AI image nodes packages

Screenshots

👇 Install community pacakges

👇 Switch workflows quickly

Development

install pip package in editable mode

pip install -e .

start nozyio server

nozyio --allow-cors

Start web dev server with hot reload

cd nozyio/web && npm run dev

Go to the web server url

Build and publish to pypi

if you haven't installed twine: python -m pip install --upgrade twine if you haven't installed build: python -m pip install --upgrade build

To build and publish to pypi:

chmod +x build.sh

./build.sh

Credits

This project is inspired by ComfyUI. Lots of code are referenced from ComfyUI (https://github.com/comfyanonymous/ComfyUI) Sincerely thanks to the contributors of ComfyUI!