node-red-contrib-tf-model 0.1.12

a custom Node-RED node to retrieve/execute a Tensorflow model

npm install node-red-contrib-tf-model

This is a Node-RED custom node which is used to load TensorFlow models and perform inference. Currently, it only supports the web-friendly JSON model format which is mainly used by TensorFlow.js. The SavedModel format will be added soon.



This module requires @tensorflow/tfjs-node as a peer dependency. You need to install it within Node-RED manually. TensorFlow.js on Node.js (@tensorflow/tfjs-node or @tensorflow/tfjs-node-gpu), depend on the TensorFlow shared libraries. Putting TensorFlow.js as the dependency of a custom Node-RED node may cause the situation where multiple custom nodes each install their own tfjs-node module as a dependency. This causes an attempt at loading multiple TensorFlow shared libraries in the same process, which subsequently causes the process to abort with a protobuf assertion error.

Install @tensorflow/tfjs-node:

npm install @tensorflow/tfjs-node


If you are planning to run this node-red-contrib-tf-model node on a Jetson Nano or Raspberry Pi 4, note that the latest @tensorflow/tfjs-node does not yet support the ARM64 or ARM32 architectures. Here are the instructions for installation:

@tensorflow/tfjs-node on Jetson Nano and Raspberry Pi 4:

  • Run the following command to install tfjs-node:
    npm install @tensorflow/[email protected]
    npm rebuild @tensorflow/tfjs-node --build-from-source

Install this module:

Once you have installed the peer dependency, you can install this module:

npm install node-red-contrib-tf-model


You can see the tf-model node in the Models category, like this:


Then you can use the tf-model node in your flow. It only needs one property: Model URL.

Config Node

The Model URL should point to a TensorFlow.js model which is in web-friendly format. Typically, it should be a model JSON file. After you specify the Model URL and deploy the flow, it will fetch the model files, including shard files, and store them in ${HOME}/.node-red/tf-model directory. You can also use a model from the local file system (e.g. file:///home/mymodel/model.json). The new node will load the model and maintain the cache entry. You can specify the Output Node name when running model inference. By default, it uses that last node as the output node.

Data Format

When performing infernce using a TensorFlow.js model, you need to pass the corresponding msg.payload to the tf-model node. The msg.payload would be a tf.NamedTensorMap object containing all the needed features in the tf.Tensor data type for the model. When a model is loaded, the input node list is output to the console which can help you to build the input named map for the model.

Example input to tf-model node:

  payload: {
    image_tensor: image

In the example above, image_tensor is the input node name of the model and image is a tf.Tensor object.

By default, the prediction runs through the whole model graph and returns the final output of that last node. You can use Output Node to specify a different node as the output node. After model prediction, results are passed to the next node in msg.payload. It could be a tf.Tensor or tf.Tensor[].


We have provided some example flows under the examples folder. They may help you to understand the usage of the tf-model node and other Node-RED custom nodes we provide.

Node Info

Version: 0.1.12
Updated 11 months, 1 week ago
License: Apache-2.0
Rating: 5.0 3




28 in the last week


  • tf-model


  • tensorflow.js
  • model
  • node-red