Skip to main content

Node.js Example

View this page on the UpscalerJS website

Demonstates how to use UpscalerJS within a Node.js context.

Open in CodeSandbox.


Tensorflow.js publishes two different libraries for Node.js, depending on whether GPU support is required.

Similarly, UpscalerJS provides platform-specific builds that parallel Tensorflow.js's platforms.


Technically, importing @tensorflow/tfjs is supported on the server, but importing upscaler alongside it is not supported. For Node.js support you'll need to use one of tfjs-node or tfjs-node-gpu. If support for tfjs on the server is important to you, open a feature request!

In this example, we'll be using the Node.js CPU platform.

Ensure we load UpscalerJS via upscaler/node, not upscaler:

const tf = require('@tensorflow/tfjs-node')
const Upscaler = require('upscaler/node') // this is important!

const upscaler = new Upscaler()
const image = tf.node.decodeImage(fs.readFileSync('/path/to/image.png'), 3)
const tensor = await upscaler.upscale(image)
const upscaledTensor = await tf.node.encodePng(tensor)
fs.writeFileSync('/path/to/upscaled/image.png', upscaledTensor)

// dispose the tensors!

Like the browser version of UpscalerJS, the Node.js version will make a best effort to handle any input we throw at it. The list of supported inputs includes a string to a file path, a Buffer, a UInt8Array, or a tensor.

By default, UpscalerJS will return a tensor when running in Node.js. We can change this to return a base64 string by explicitly specifying the output:

const tensor = await upscaler.upscale(image, {
output: 'base64',

Next, read about how to specify a custom model when running under Node.js.