A new release of Topaz Video AI is now available.
Released October 10th, 2023
Changelog from 3.5.2:
Adds “Black and White” toggle to video input options.
- This option will desaturate output to prevent models from adding color to Black & White videos.
Adds “Telecined” toggle to video input options.
- This option will reverse the telecine process and deinterlace your video to 23.976 fps before applying AI models.
- Adds detailed tooltips to Output Settings.
- Enables ProRes, H264/5 and AV1 encoders for Linux.
- Fixes many color-banding issues seen in the in-app player.
- Fixes update issue related to frame interpolation / ffmpeg that some users experienced after updating to 3.5.1 and was not fully addressed by 3.5.2.
- Fixes tooltips so they never go off-screen.
- Fixes UI issues in video settings when information displayed is multiple lines.
- Changes filter tooltips so they are now also seen when you hover on the filter name and not just the icon.
- Changes “os” field in video-encoders.json to a regular expression (regex) evaluation to give more flexibility.
- Currently not all the models are supported on Intel ARC (model support will be added week to week).
- Videos with mismatched metadata and streams will display incorrect duration
- Frame number preview length may shorten on app restart
- Using Telecine mode can cause a mismatch in playback over time. This will be addressed in a future release.
- B&W toggle is causing a color shift in image sequence output.
This does not mention the issue with upsidedown or sideways output videos?
Is there a sub-forum for enhancement and feature requests?
I don’t know where to post this, but can you consider adding audio/video muxing feature that let user choose an alternate audio to use with the video? This will be a very useful feature because when upscaling 240p or 480p, the audio in the source video is usually in poor condition. They required some editing such as amplification, channel mixing and normalization to bring up to good condition. If there is A/V muxing feature, users can edit the audio in external editor and select the resulting audio file in topaz, upscaling and audio muxing in one step. Without this feature, I have to mux after upscaling which reduces video quality.
I think I reported it during the beta phase and also sent my log files to the initiator of the original post as suggested for privacy reasons but
- I got not a reply
- It was supposedly not addressed
- the issue persists
The application crashes/disapprears unexpected - which did not happen to versions before - and the ffmpg continues running alone. To prevent the log files to get lost via private message again and because a dropbox is not told I add my log files here.
logsForSupport.zip (249.2 KB)
What container are you using? There should be no change to the video when adding audio back in.
I use mkv because it can hold multiple audio tracks, video tracks, subtitles, chapters and everything else. And it plays on my Roku player through Plex. I use GoldWave to edit audio and VidMore to convert it to DTS or AC3. Once I have it how I want it, I join it back into the mkv container with MKVToolNix. No change is made to the video.
I agree some audio editing tools would be nice in TVAI, but that’s not what it’s made for.
We fixed an issue where videos with some specific rotations/strings were opening upside down. Please let us know if you’re still running into this. Thank you.
What about Iris being abnormally slow on Apple Silicon?
Is that going to be fixed some day?
For any feature requests/ideas, please post on Ideas - Topaz Community. Thank you
Little question: when will there be a template for text, or will it be included in the Iris template? Because in SD to HD videos, scenes with text (e.g. on books, newspapers, billboards, other text on objects or buildings, etc.) are not always easy to read. It’s not always easy to read on the SD original, but on the HD output, it’s even worse most of the time. It’s totally unusable. In fact, the output text looks like gibberish. Just like in Photo AI before the introduction of the text reconstruction model.
You can replace an audio stream using FFMPEG without re-encoding the video or audio…
ffmpeg -i video-subtitles.mkv -i audio.aac -shortest -map 0:0 -map 0:1 -map 1:0 -codec copy output.mkv
Change file extensions as you wish.
Note “-shortest” trims the video at the point where the shortest file ends and isn’t required.
Assumes the first two streams in “video-subtitles.mkv” are the video and subtitles. If the original audio is there change map 0:1 to 0:2 or more.
If you only want video from the first file just have “-map 0:1 -map 1:0” (without quotes).
If you want more streams from the first file add “-map 0:3” etc., for the second file “-map 1:1” etc.
I don’t need audio editing tool in TVAI. I can do that in external audio editor like Cool Edit Pro or Adobe Audition. I just want TVAI to add an audio option that allows me to select the audio file that I edited in external editor. I will explore MKVToolNix, but it would be nice if TVAI has the feature built in.
Thanks. I will take a look at the ffmpeg.
Video AI is actually doing all the work using a modified copy of FFMPEG, or you can get freeware builds for Windows (without the Topaz stuff, so no licence required on the system it’s running on) in the “Release Builds” section here:
i.e. file “ffmpeg-release-full.7z” which is the version I’m using (6.0).
From the archive you just want ffmpeg.exe from the builds directory. (Also possibly 7zip to extract it.)
I don’t think it’s necessary.
Don’t try to make TVAI powerful video editing software, let’s concentrate on improving what’s it good for - video enhancement.
As per your request - you can use a ton of free muxing software to mux final audio and video without re-encoding.
The Win release (along with the previous one) is generating immediate errors on output files whenever the ‘Include Live Preview’ option is turned off on the Output settings, despite full re-clean installs (including the deleting of known Topaz related registry entries and ProgramData subfolder).
I did raise this in the bug report thread but as yet no one has responded so just putting it here too in case it hasn’t been spotted.
After updating to 3.5.3, in settings there is an empty option when I choose GPU, and if I choose “ALL”, it won’t processing and gives an error.
Does any one meet this after update?
The way AI works, if it were to make unreadable text readable, it would always replace the original words with whatever words were in the training data. Reading it would be as pointless as trying to read the unreadable words.
For words that can be made out and read, some models do make them worse or unreadable. It would be nice if they only ever made them better. I think Gaia is pretty good at leaving them alone when it cannot make them better.
The slow motion drop-down is no longer editable. This could be a breaking change for some people.
This is not an issue for me because I use a script to generate the ffmpeg commands, but it can be an annoyance when I want to do some proof-of-concept work in the GUI. Basically, I always use a slow motion factor of 2.5. Now I have to modify the TVAI generated commands just to get that back in. Even if I want a simple quick preview.
Won’t allow me to use a custom resolution using prioritise by scale. Back to 3.5.0…again.
Unfortunately, version 3.5.3 still has the same errors with processing MPEG2 576i files… I am attaching the logs.
2023-10-11-20-11-10-Main.tzlog (91.3 KB)