They already have to solve many problems before with the VEAI, because they would create new ones as a product that, as mentioned above, doesn’t even exist yet, anyway…
Att.
Druid®.
Yeah, you’d think so. I feel like I’m taking crazy pills when updates continuously release without this simple solution FINALLY implemented. Oh well.
I learned early on to never use VEAI audio pass through, because it sucks. I’ve been using MKVToolNix since day one to put the original audio track and subtitles back into the upscaled video. Only on a few occasions have I had synchronization problem, because VEAI was dropping frames, but not very often.
1 Like
The reason im asking is because the audio would still intact when im using 1.8.1 version.
After upgrading to 2.2.0, every output i made loses its sounds eventho i turned on the “keep audio” option.
So in summary, theres nothing wrong with my setup but in 2.2.0 they purposely removed the audio in the output video?
If yes, why would they still keep the “keep audio” option in the UI eventho the result will always have no audio?
No, they did not. It has always been broken and remains broken. Sometimes, in some situations, it works, but it has always been better to use an external program to handle the audio.
What… so why the earlier version im using has no problem with the audio output?
With this version, everything i made have no sound. I tried like 10 different video with different format and specs but still no sound. I thought theres something wrong with my setting.
Im thinking downgrading to earlier version if this only happened in recent newer version.
I can’t speak for version 1.8.1, but 1.6 and 1.6.1 had all sorts of audio issues.
1.6.1 VEAI would convert mono-channel audio to dual-channel and convert 5.1-channel to dual-channel when it did not crash before finishing the upscaling. If the video had any sort of issue in the tracking, it would throw off the audio tracking when remerged with the original audio.
What I finally did to get 1.6.1 to work the way I wanted was to use FFmpeg to create a new CRF 0 libx264 mp4 video with no audio and feed that into VEAI to upscale. That would work around all of the tracking issues and prevent VEAI from crashing when modifying the audio.
Finally, once I had a video upscaled with VEAI, I used an FFmpeg bash script to merge the upscaled video back with the original audio and any subtitles to make a finished mkv.
No idea how downgrading would help, if 1.8.1 is anything remotely like 1.6.1 when dealing with audio issues. My $0.02, if you are not comfortable with FFmpeg, use MKVToolNix to merge back the original audio into your upscaled video. It should only take a few minutes at most to merge the original audio with the new video into a new file.
The bottom line is, you don’t want the audio output from VEAI. Ever. Not until they fix it.
I use 2.1.1 (rtx3090 i710700). “Keep audio” works fine (in stereo mode, without subtitle) with all the models, except the interlaced Dione, where there is no sync … and it is useless.
I suppose that’s fine, if you are upscaling family home videos with basic LR stereo, but forget it if you want to keep the original 7.1 Dolby Atmos in the upscale by using the “keep audio” option.
3 Likes
Why output model usually wrong from what I picked???
There’s an annoying bug now where if you change from Low/Medium/High, it will show the recommended model, but it won’t actually change your active model. You need to click the dropdown menu to select it again. If you import a vid, and the default is GHQ, if you swap to for example Medium preset, it will show AQ at the top, but the applied model is still GHQ, you you need to manually select AQ again. Your active model is displayed on the bottom right.
This really needs to be fixed.
For people using Nvidia cards (mine 1080Ti) , I was getting random lock-ups and after looking into it for some time, I found the Nvidia HD Audio device driver was the problem - i’m not the first to have this issue.
I disabled it in device manager (because uninstalling it just comes back after reboot) and PC has been on for 6 days! Couldn’t get more than a few hours out of it before that. I re-enabled it and lock-ups just come back.
I now need to see if I can use optical for the sound now!
Send a message with Taylor Bishop tagged in it. There’s only a day or so before the next release.
Perhaps the developers or the Topaz Team responsible for VEAI could consider an idea that came to me yesterday (yes, it is insane, but I thought it couldn’t hurt to ask):
How about (if that were possible) adding a sort of “Genre Templates” that users could add to existing models to adjust them more specifically? What I mean by that: The already existing models have been trained on a variety of “basics” (like people, faces, objects and such) and on different source quality levels (low res, high res, with artefacts and such), but what about creating templates according to genres, that will “tell” each of the models to more specifically address issues that are source dependant, no matter what model has been chosen - a couple of examples that I thought of yesterday:
- Template “Action”: Focuses more on fast moving visual elements and avoid tearing/smearing and/or problems with lines/edges
- Template “Sitcom”: Focuses more on people and faces (close-ups) with indoor scenes
- Template “Fantasy”: Focuses on the on-screen fusion of real life action and cgi/vfx elements (like blend animation or green screen and such)
- Template “SciFi”: Focuses on dark scenes with difficult lighting and vfx/cgi or model shots with rainbowing around edges
- Template “Anime/Cartoon”: Focuses on lines and edge preservation as well as block artefacting and banding on large color areas
These could be done as a sort of “buy-in” option, so people could choose which templates they want; you could do a trial period for people to decide, offer discounts on template packs with templates from similar areas of interest (like SciFi & Fantasy) and so on.
Just a thought, that way to could financially compensate the work involved.
Now that an official Windows 11 build is released to insiders, I would be able to test VEAI as a beta tester. 
yeah, you just remux original audio to upscaled video with mkvtoolnix. Doing that also helps you to complete the video, which is sometimes messed up by VEAI. The file output by VEAI is somehow usually troublesome, either .mp4 and .mov
Interesting idea. You should really post it in the ‘Video Enhance AI Feature Requests’ topic in the Beta Testing section.
Hi all,
A new version of Video Enhance AI is available. You can view the release thread and download the new version here.
1 Like
