VEAI New Features Request

  1. Adding “Open with Video Enhance AI” to the Windows 10 Context menu (right mouse click button)
  2. Adding ability to open VEAI from other software, in example Fast Video Cataloger (External Video Editor) works with VirtualDub but not with VEAI.

VEAI%20Windows%20Context%20Menu Fast%20Video%20Cataloger%20Preferences

I’m wondering what the balance between spatial and temporal denoising are for the various algorithms.

I like how Theia allows us to adjust settings for sharpening and deblocking, but as far as reducing noise goes, I’d like to adjust settings based on spatial and temporal elements to denoising. Often times Artemis blurs too much when there’s enough grain and a fair amount of motion, but I really like its overall detail recovery - hopefully we can see a new setting for bias to temporal denoising over spatial.

So far, VEIA is doing a good job denoising, sometimes upscaling, sources already of a decent to high quality. I would love to see two new models emerge: one trained specifically against low bitrate videos to get rid of blocking and other artefacts introduced by high compression (as it stands now, these artefacts are simply upscaled alongside the rest of the video, no matter what model I choose); and one model trained specifically for deblending (I already proposed that but I thought it couldn’t hurt to have a trace of it in this feature request thread).

Hello. I’ve been experimenting with this software and I’m very impressed across the board. That said, I have one feature request that might make a world of difference.

When using some of the more impressive settings (Artemis) to my eye, details inherent in skin seem to take a bit of a hit while processing (freckles, beard stubble, imperfections, etc.). When these elements get smoothed out, they move towards creating a slightly unrealistic look on human subjects. Yes, this is subtle, but important.

My feature request is some kind of slider that might allow us to vary this one smoothing element of the software. and perhaps a second slider that balances where this micro management takes place, to protect dark areas that might benefit more from the smoothing (removing noise etc).


1 Like

pls, mts file support.

Pls, add option to e encode and decode video by gpu.

It would also be really cool to have a frame interpolation algorithm, to double framerate, for instance.

1 Like

Greater use of gpu, at least as an option. Now only Gaia uses a little more gpu.

Facial recognition. AI must recognize a face. This is very easy in 2021. But a face must always have a human appearance! May be not 100% fidelity, but 100% human!

Dione should be able to recognize for himself the correct order of the frames. This doesn’t need a lot of AI.

8k mp4 encoding capability without crash. Only MOV is a limiting in use.

ability to pause the process

1 Like

It’s all and well that you can preview the results of a specific ai model, but it’s not very useful without being able to compare different models side by side for a specific video to see which one gives better results.

So it would be great if instead of just comparing to the original you could compare two different models results in the preview.

That’s already been through testing and is on the way for a not-far-off release.

Sometimes older video clips have an incorrect aspect ratio set on them, it would really speed up the workflow if we could fix this when running video enhance ai. In so eliminitng the need for an intermediate step in another application.

Also it would be great if we ware able to apply custom amount of crop to each side of the video seperately. For example VHS transfers have variable pixels of garbage on each side of the frame.

1 Like

Many people have DV or VHS recordings. Running Topaz AI on some downloaded videos does not make sense, because better source can be found.
Your customers are people who want to save their home videos.
DV format has specific encoding, jpg compression artefacts etc. Can the training be DV specific, to get the most of the format?

agree, so i suggest to have an option to see the result for all models on one screen - to easily decide, whats best. You will select a button - “compare the models” and then one video will be created with screen divided to regions, each of them done using different model.

This comparison of models is currently being done in the VEAI 2.2 beta, but it’s slow and clunky.

VEAi has gotten really good!

With some video files moving the Position Indicator is too sensitive, and it is hard to move by a small number of frames, while at the same time using the Frame by Frame positioner (or L or R Arrow Keys) is way too slow.

It would be great to have a way to move faster but with more control over the speed. Perhaps holding the Ctrl Key while using the L or R Arrow Keys or the Frame by Frame controls.

Basically three new features I’d like to see:

  1. Some sort of OCR added to the algorithm so that diagrams, charts and credits are more legible.
  2. I’d like to see the AI take the two preceding frames into consideration and interpolate a progression as it calculates a new frame. This way there should be a smoother transition between frames and less flickering, particularly with Gaia.
  3. A built-in or plug-in or external reference to a colour and levels adjustment system, such as Davinci Resolve, to be able to do all the processing in one run.
1 Like

I’d like to see a merge video function , so I end up with one file. Also there’s an error with D tv3 de interlacing with double frames, it ends with a file with audio mismatch, like the audio is stretched

1 Like

I’m a CG Artist currently testing 3Delight which doesn’t come with a built-in denoiser.
I know there’s Intel Open Image and Nvidia Optix Denoiser out there, but the result of these 2 often softening the final result too much.

I’m wondering if VEAI has any plans to support 16bit and 32bit multi-channel .exr image sequence? so that we can denoise all of the contribution AOVs in one go? and also would be great to have a batch render option to denoise multiple render layers simultaneously.

Hi yall an update on how to submit feature requests here: