Bundled browser image upscaling dependencies:

TensorFlow.js 4.22.0 — Apache-2.0.
Source: https://github.com/tensorflow/tfjs
Bundle: https://cdn.jsdelivr.net/npm/@tensorflow/tfjs@4.22.0/dist/tf.min.js
License: tensorflow-LICENSE.txt

UpscalerJS 1.0.0 — MIT.
Source: https://github.com/thekevinscott/UpscalerJS
Bundle: https://cdn.jsdelivr.net/npm/upscaler@1.0.0/dist/browser/umd/upscaler.min.js
License: upscaler-LICENSE.txt

@upscalerjs/esrgan-thick 1.0.0 — MIT.
Source: https://github.com/thekevinscott/UpscalerJS
Models: https://cdn.jsdelivr.net/npm/@upscalerjs/esrgan-thick@1.0.0/models/
Definitions: dist/umd/models/esrgan-thick/src/x2/index.min.js and x4/index.min.js
License: esrgan-LICENSE.txt
Modifications: Model weight files were renamed from group1-shardNof7.bin to
weights-N.data. Manifest paths were updated accordingly. Weight bytes and
model topology are unchanged. Both 2x and 4x models are bundled.

The optional gentle noise cleanup and edge sharpening are conventional
pixel filters, separate from the neural network upscaling step.
Photo and drawing modes use the same pretrained model; filter settings differ.
This implementation does not include a face reconstruction or OCR model.

@upscalerjs/esrgan-medium 1.0.0 — MIT.
Source: https://github.com/thekevinscott/UpscalerJS
Models: https://cdn.jsdelivr.net/npm/@upscalerjs/esrgan-medium@1.0.0/models/
Definitions: model-medium-2x.js and model-medium-4x.js (dist/umd x2/x4 index.min.js)
License: esrgan-medium-LICENSE.txt
Modifications: group1-shard1of1.bin renamed to weights-1.data; manifest path updated.
Used as the fast model for larger images (about 10–15× faster than the thick model).