Video Enhance AI v1.8.1

I’ve heard people say there’s a point that the human eye can’t tell the difference, so 0 crf is a waste, thoughts?

Also is exporting to PNG be the same as 0 crf mp4 since it says lossless?

I certainly can’t see the difference between 0 and a higher number, but a lot of people here see problems that my eye does not pick up. The default for h.264 is 23, and 17 was apparently chosen as being so good that lower doesn’t matter.

Considerations: CRF 0 or lossless PNG will result in extremely large files, larger than many of us can handle unless the video clip is very short. A lot of computers might have a problem with keeping up if you tried to play a lossless file. On the other hand, if you plan to do many steps of post-processing, starting with lossless would be an advantage, only in the last step doing compression and becoming lossy.

This is one of a thousand things about working with graphics where just about everybody will have different goals, and different tolerance for perfection/imperfection. As I said, my eye is apparently not all that good, so I’m quite happy with moderately lossy output.

Don’t forget that MP4 h264 is 4:2:0 chroma subsampling even at CRF 0, PNG is RGB so it will produce better outputs with more color fidelity and maniability for further editing.
Real lossless in VEAI is only still images, PNG or TIFF.
And that’s why you better choose ProRes 422HQ instead of H264 CFR 0 if you don’t want to deal with images.

Does anyone know what’s the equivalent in CLI for 4KUHD with “Crop to fill frame” preset in GUI? I’m trying using combinations of scale, dimension, and fill mode flags, but I’m not getting what I can get with the GUI version. My input is 640x480 with black bands on the top and bottom. Specifically, I want to use 6x scaling and output should be 3840x2160. As scale and dimension flags can’t be used at the same time, there’s no way to control aspect ratio for scale; thus dimension seems the way to go, but it computes a scaling factor based on height, which results in 4.5x, making black borders around the image. Should I use 1) 6x scale and remove borders using external tools for pngs, 2) crop video to 480x360 and then apply VEAI, or 3) is there a way to achieve only with the given CLI? Thanks for your input in advance!

The PNGs aren’t that bad…it’s the TIFs that are ginormous.

Developer. This is wrong. Leave all models in the directory and only delete those which
are incompatible (ask about all during update/install).

Nobody wants to re-download it all again.

Hope the build-in deinterlace will arrive soon.

1 Like

I am currently remastering the Stargate SG-1 Season 1 to 7. I’m at 0.16 seconds per frame as on the screenshot.
Lien vers mon image
Sorry I can’t do better for uploading my images. The site doesn’t want in direct link mode.

Hi, we are very aware that you do not want to have to download a file that you have already downloaded.
This has been made abundantly clear.

We are in mid development of addressing this with a feature.

The feature will be released when it is ready. We have many other factors to consider in our current schedule.

We do understand that you are all very eager to have this feature and we are working as fast as we can to get something out for you so that you do not have to download files again that you have already downloaded.

Thank you for your feedback.


Lots of more important things need to be addressed first.

@tabnk @lee.keels

Please take into account that what is important to one user, may not be as important to another user and vice versa.

We need to hear about what is important for all users .

It is just important for us to hear about a user desire for deinterlacing as it is to hear about a feature for a feature in the installer to reduce downloads,

Basically, we would like to hear all perspectives about what users are interested in. Thanks.


One-shot model download option would be nice. Download all available models and block sizes for your system/GPU. Then, no more waiting for models to download while you are trying to upscale. You could even work offline with full efficiency, having all the fast models available.

Something that could be added, would be to see the time already elapsed when rendering.

I think those already are in 1080p.
I was going to start doing this yesterday, but checked if they were released. But since they already were in 1080p, i started upscaling the Babylon 5 movies. :stuck_out_tongue:

Sure. Artemis LQ removes details (especially ‘noises’) and create new details. Final result is absolutely fake. But the sharpness and produced details are AMAZING, that’s what an AI could do.

So maximum output for the AI feature is 400%? Also, Spline > Lanczos hehe.

Gaia HQ + Artemis MQ + Spline upscaling. Ahihi.

On 1 pass, yes.

For downscalling, usually yes. But you have to tell that to the devs, not for me

That’s what I’m doing in 1.8.1. I prefer GAIA AFTER Artemis because it also adds ‘granularity’ over artemis that looks too soft for me