Video Enhance AI v2.2.0

Thank you for your contribution. I shall look carefully later, and attach all previous attempts to show you different results. Probably there are users who have some Ful HD 1080 footage so it could be helpful for them to choose right denoising method. I wonder if different software like DaVinci or DVO Clarity could be any better/worse.

Is there a guide that we can all use to make sure we don’t have these ā€œmismatchesā€?

Not sure about changes in color, but veai have been messing with the gamma value of all upscales for quite some time now which possibly might give the illusion of shifting in color/similar.

I personally use this avisynth filter: Levels(0,1.02,255,0,255) in staxrip/hybrid in the process of encoding my img sequences to remediate, but you can use any capable software to raise the gamma levels by 0.2 if you want.

Hope this helps!

1 Like

Ok, nevermind, nevermind.

I have just found out that the oled screen on my phone has a massive green to magenta color shift from top to bottom, which I was correcting for unneccesarily.

Welp that sucks… :frowning: Hope you can get it repaired.

1 Like

Question would this software be able to upscale my 1080 10-bit 4:2:2 120fps footage from the R6 to UHD and maintain the 10 bit? I would like 10 bit so I can push the colors more around without it falling apart so quickly. Will the upscaled version lose the ability to color grade it like ā€˜normal’ 10 bit footage? Does it turn into 8 bit which is limited in color grading abilities.

PRORES 422/HQ are 10-bit AFAIK so, there you go.

After a week or so with this, @taylor.bishop I’ve noticed the following oddities.

  • Using a M1 Mac mini os 12.3.1 when the machine puts the ā€œDisplayā€ to sleep, but the option to ā€œPrevent computer from sleeping automatically when the display is offā€ is checked in macOS Energy Saver System Preferences. VEAI slows down to 1/4 the ā€œawakeā€ speed, and shifted a 23hr render, to being nearly 56hrs because I was gone from the house for 2 days and never woke the screen. Is there something that changed, why would the display being off cause the machine to act asleep in regards to VEAI? (no other app does this in regards to large downloads, iMovie rendering, Handbreak encoding, etc, so the machine is awake, just the display is asleep.)
  • When selecting a Save Directory from VEAI, and the Finder Open/Save dialog displays, if you are in a directory that you want to change, and you submit that directly, it will replace the ā€œVolumes/HD/Folder/Subfolder/ā€ with just the ā€œ/Subfolder/ā€ giving now save directory info. To fix this, you must click on the ā€œ/Volumesā€ side bar, then ā€œ/HDā€ > ā€œ/Folderā€ > ā€œ/Subfolderā€, every time you want to change the directly in Preferences. As though the detection tool is no longer receiving the ā€œCorrect Directoryā€ information when you begin to change it. Make sense?
  • In Comparison View it appears you are unable to ā€œdeselectā€ a comparison model. So when you begin, you have ā€œOriginalā€, ā€œFirst Modelā€, ā€œSecond Modelā€, but you must click, and specify the ā€œThird Modelā€ comparison. Is there a way to only have 2 comparisons, and return the ability to ā€œclick show originalā€ when comparing those 2 models. So a split screen left and right for 2 model comparison, and us to click either of the models to display the original video resolution underneath the same way you do in the single render option. Depending on the video, rendering can be slow, and I’m down to 2 comparisons, it would be nice to see 50/50% screen space those 2.
  • The option to ā€œQuitā€ has been removed now if an ā€œActive Renderā€ is in place, thank you, however if you are using the App Switcher ā€œCommand+Tabā€ on macOS, if you accidentally hit Command+Q, while in the tab switcher, it will exit the VEIA app without notice, and could lose hours of encoding, as it has for me a couple times. An application prompt ā€œif and only ifā€ an active rendering is in place… and allow us to Quit the app without a prompt, if only videos are no longer rendering. (While I really like the flexibility if exiting without saving, I felt that an active 20+ hour render should or any render, except for previews, should prompt a warning. ā€œActive render in progress, are you sure you want to exit?ā€
  • Lastly, would there be a way to purge old preview renders from RAM/VRAM space, because if I preview compare, ore just single preview many different sample areas, and settings, the VRAM portion has become 48GB, and 56GB, and has crashed when in the middle of a long render, unless I restart the app before beginning the actual render, to clear VRAM first.

Great upgrade, keep it up!

1 Like

I have collected different methods and prepared the compilation. Any comments, preferences? Of course I do not expect miracles from such low light footage but in general | have a lot of footage recorded in low light, some of course not under such poor condtions and would like to use 1-2 methods to clean footage.
https://we.tl/t-HHQEUBzM2I

Reporting a bug. (I posted this into Facebook group first.)

I discovered a bug in 2.2.0.3b where it seemingly at random processes the same frame twice and skips one frame. In other words, if the source had frames ABCDE, the output had ABCCE.

The source was an mp4 file, Baseline@L4.1 AVC with 4 reference frames, 1080p23.98 encoded at 10mbps. Output was set to ā€œProRes 422 (Slow)ā€.

I converted the mp4 source file into ProRes first in VD2, then fed that .mov into 2.2.0.3b, and ran a quick test. Output did not exhibit the problem.

Tested release v2.2.0 a moment ago, and the same issue is affecting it as well. Same result when testing with another mp4 file.

@taylor.bishop in 2.2.0 I am still seeing the wrong model being selected from the CLI when changing models. On first launch when using a different model than the one used last time the software was run, it will say the correct model, but appears to be using the prior one (as evidenced by the processing time being very different than expected).

There’s also always an odd display when running from the CLI, it says the model + ā€œERRā€

image

I’m working around this by scripting to start VEAI with start-frame: 0 end-frame: 1 with the desired model, then restarting it again with the full encode settings, and it seems to switch it on the 2nd invocation.

I am pretty sure I’ve observed this too. I never went and actually compared the output, but I’ve seen VEAI rapidly skip over a frame while watching it process.

I haven’t seen it discussed but are there any plans for VEAI to be able to copy over subtitles and chapters?

Maybe you should try MKVToolnix.

I’ve been using for months but haven’t seen frame skip before. Maybe because of the footage itself.

I do use MKVToolnix frequently. However, it would be FAR easier to have this copy functionality when it comes to very large batch operations like TV series. I’d really rather not have to do that process one-by-one when it seems totally reasonable to have basic copy function in-app. Additionally, I am not interested in learning to code scripts myself to accomplish this.

It might have something to do with the container, mp4 in the case. I’ve tested with two PCs and two different mp4 files so far; frame doubling/skipping in all cases. No such problems with .mov source.

I haven’t used MP4 for weeks, so I don’t know what they changed inside VEAI. But I recommend using TIF to control freely the quality and file size.

Frame skipping is a known issue and is planned to be addressed by Topaz in the VEAI 2.3 release (according to Taylor Bishop).

May Release: v2.3.0
Multiple video I/O fixes (frame skipping, frame cut-off on MKV, 4:2:0 chroma fix on interlaced source, plus others) and codec / output additions (MKV output, different ProRes options)
FPS conversion model OR Artemis / Dione model updates

June Release: v2.4.0
Better audio support (correctly copy with supported codecs / input formats, issues with audio and framerate conversion)

July Release: v2.5.0
Projects, separate Render Queue application, and Preset Manager

3 Likes

I use MKVToolNix to restore the original audio track and sub titles. For batch processing, its best to setup default values for labels and destination folder. From there, multiplexing is simply point and click.