The waifu2x on Mac.

The new version is capable of taking advantages of CPU, GPU, and ANE.


  • Enlarge videos or images with machine learning on Mac.
  • Interpolate frames for videos.


Files and source code could be found in releases.


This app works completely offline and requires no internet connection. Nothing is collected or stored, expect for:

  • Your settings stored in its containter.
  • Temp images in during comparison. (These files will be deleted when the windows is closed.)
  • Temp images during processing in its container, the existance would only last for three lines of code, after which it is deleted.
  • Output files and logs (if you turn on “enable log” or “enable dev” in preference) in its output path.

If the app crashes, please choose not to share crash log with Apple.


The models where obtained from waifu2x-caffe, and translated to coreML via coremltools.


This app was based on the work of waifu2x-ios. Nearly all the files in the folder “waifu2x-mac” were created by him. However, modifications were done to improve speed.

The files inside the folder of dain-ncnn-vulkan-20210210-macos can be downloaded here.


When processing a standard 1080p image (1920 × 1080), MacBook Pro with the M1 Max chip took only 0.7 seconds.


This app was written with SwiftUI.
Screen Shot 2021-12-12 at 2 53 54 PM


Screen Shot 2021-11-29 at 2 44 41 PM

Denoise Level

You can compare results from different model by choosing Compare > Compare Models. Example:
Screen Shot 2021-11-29 at 5 58 27 PM



View Github