Video Enhance AI v2.3.0

Download: Windows | Mac

Major Features

  • New model: Chronos Slo-Mo / FPS Conversion. Use this model to increase your video’s framerate, or make any video slow-motion
  • New model: Proteus 6-Parameter. Use this model to fine-tune deblocking, detail recovery, sharpening, noise reduction, dehaloing, and antialising
  • Preset Manager: Allows you to create and switch between multiple presets, as well as download and load them


  • New tutorial: Slo-Mo & FPS Conversion
  • Engine improvements for better speed and stability on a large range of hardware
  • Added the ability to switch between showing frames and showing timecodes in the UI
  • Processing completion time now displays in hours:minutes:seconds instead of just seconds
  • Application should resize correctly at extremely small resolutions
  • Estimated completion time and seconds per frame are now averaged out over 3 frames, making them slightly more stable
  • Playback speed in the preview is now based on output FPS if there is one, else the input FPS
  • Tutorials have been revamped to allow forward / back movement without having to explicitly click the suggested buttons


  • Fixed excessive memory usage when switching models
  • Fixed a memory leak when exiting preview mode. This will have a major affect on memory use when previewing, especially when using Comparison View
  • Fixed a frame-skipping issue when previewing the deinterlacing models that double the framerate
  • Fixed .mp4 getting appended to .mov files when choosing a custom output location, which was causing write errors when processing
  • Feedback dialog now allows click-to-close and has been added to the Mac Help menu
  • Menus now have borders around them so they no longer blend into the background
  • Fixed keyframing tutorial not selecting the correct model when starting the tutorial
  • Grain Settings now get applied to all models when using Comparison View (previously, they would only apply to currently selected one)
  • Model parameter sliders (for Theia, Proteus) now allow you to click and hold to change their values
  • Comparison Preview panels should now pan correctly when the preview is rendering or looping
  • When adding videos to the list when videos are already open, the first one added will now be selected
  • When processing, the playback controls now disappear and are replaced with a frame counter showing how many frames have completed processing
  • Removed the ability to launch the application after install, which breaks drap & drop on first launch
  • Added a cap to the amount of frames that can be previewed
  • Preference panel is now scrollable, and should no longer overflow at extreme resolutions
  • Preference panel should now be faster to load
  • Better error messaging when path is too long on Windows or an invalid video is being previewed or processed
  • Application will quit processing a video that it cannot write to

Best update ever. Thank you very much!


Be sure to also check out the Product Update Post for this release which has a bunch of helpful tutorial videos and other clips that showcase all of the powerful new features in Video Enhance AI!


When will this version be listed in downloads on your website?

1 Like

It’s been updated now, thanks for the heads up


How does the Chronos AI model compare to RIFE neural network frame interpolation for increasing framerate? I’ve been using RIFE via the Flowframes app.


Just Uploaded Files for Noise Section.:slight_smile:

In general, RIFF is faster doing 2x/4x, Chronos produces better quality and is flexible for any scale up to 20x. But they all depend on the input videos.


@taylor.bishop New Model Unable to clear Noise From Video even at 100%

It doesn’t look like you’ve run the preview.

That’s not noise, that is half the pixels of an image. lol

It looks more like a artistic style than digital noise. There is no chroma noise only luma noise and the grain size is the size of the fist. That does not look like a typical noise patter on digital camera, so I imagine it is not treated as one. If you want to fully get rid of “noise” you would either have to reconstruct 50% of the image, or go the old school way and just apply blur to the image. But I would not use this as the test image for noise reduction is all I’m saying.


That is literally the noisiest footage I’ve ever seen (you might as well be upscaling TV random white noise). You’re expecting too much, they’re not gonna train the AI for unrealistic noise levels like that.


The Luma Noise is Static .Low Quality Video Setting Removed the noise Nicely The New Model with Adjustable Could noT.


Proteus looks very impressive in the video demo. I was not able to generate as much texture detail as in the tiger video, using the beta. Was there improvement to the model between the last beta and this release?

There was. We typically try to give the models as much time to train as possible, and will package them at the last minute before a release. So this version has about a week’s worth of extra training compared to the last beta version.


You get my samples ?

Bishop, how about to enable the application to get additional trainings data when used at the users end? That could safe a lot of trainings time on the companies end and improve the trainings data for all continously while it is being used at the customers end. :slight_smile: