I don’t expect you to have read all the posts before. This is because the generated command sets the scale to 0. If you copy the command and change that 0 to 2, you will get the exact same result as what you get in 2.6.4.
In my opinion, 3.0 is still beta in ‘release version clothing”. But, the premise of the new approach to the program is a good one, and it is destined to be much better than 2.6.4, in time. I would give it a few months to get there.
1 Like
What you say makes sense. Question is, is it this way in error or intentional on Topaz’s side? Have they commented on this point yet?
1 Like
No, they have not. There are only a lot of posts in here. Maybe if I make a new topic outlining the whole thing I’ll get a reply. I also don’t know if it’s true for every base model, but I tried generating a command for the specific model and scale mentioned by @Chase and it did indeed set the scale to 0 instead of 2.
1 Like
Golf clap, I’ve attempted to update from v2.6.4 about 5 times now, VEAI downloads the update package, seemingly applies it, but opening VEAI I’m still on v2.6.4 with update prompt. Youre just wasting my data man, tired of the bugs and bullshit with your software.
1 Like
Stop trying to update from v2, uninstall it, and install v3 from the links at the top of this post.
1 Like
What? Copy command and change to 2?
Are you talking about the command line interface?
So I set the scale to 2x, but, it doesn’t scale it properly unless I run a command line?
And this… is the final published version of the program?
1 Like
Whoops, I must pay another $300 to update as my original version expired in May of this year. Trying Dione and Gaia, I see no improvement with the main problem with VEAI, that it desatures the image, I’m talking about how it pulls out crimson sub-surface scattering from skin-tone, so I won’t be paying the $300 update tax. Golf clap once again.
Yes.
Be very careful when installing 3.0:
- Be sure to specify a different location for the software
- Be sure to specify a different data/model location
- Be sure the checkbox is clear to uninstall 2.6.4.
They will co-exist fine.
2 Likes
I just tried running a Video AI encode from the command line (using the copy ffmpeg export command).
I ran it in the Video AI encode install folder and got this:
[swscaler @ 0000020CD2C8C5C0] [swscaler @ 0000020CD2E23D00] No accelerated colorspace conversion found from yuv420p to bgr48le.
[Parsed_veai_up_0 @ 0000020CD15C4C40] Failed to configure output pad on Parsed_veai_up_0
UPDATE: I see the problem, environment variables aren’t set (didn’t open command prompt from Video AI)
@Topaz
Your ffmpeg command line building is quoting a lot of things that don’t need to be quoted.
"-hide_banner" can be just -hide_banner.
Also:
-vsync is deprecated. Use -fps_mode
Passing a number to -vsync is deprecated, use a string argument as described in the manual.
1 Like
This is me just guessing, but in 2.6.4 setting the scale determined what version of the base AI model to use. (Like from DVD resolution to FHD or 4K or what not.) They seem to have implemented that same logic in 3.0.0, but left out that part when generating the command. For example, here’s the offending section of the command:
..."-filter_complex" "veai_up=model=alqs-2:scale=0:... should be:
..."-filter_complex" "veai_up=model=alqs-2:scale=2:...
They at least tried to make it easy to run it from the command line. Says I can press Ctrl+t to open it. Then in the output panel I can click the […] button and select FFmpeg command. Then I can copy that and paste it directly into the command line or a text editor and change it. (Oh, you might have to set an environment variable…)
Final is a strong word here. Everyone who doesn’t want to play with the command line is doomed in that case. I’d give them a few more months to release what was actually supposed to happen with this version.
I am suffering 2 issues with version 3.0.0:
-
Audio saving. When processing video with multiple audio tracks (e.g. english, german, comments, etc), only first audio is being copied/transcoded (the rest are discarded). The version 2.6.4 used to copy all audio tracks, saving appropriate track names in the metadata too.
-
Making the bitrate dropdown rather than quality level for H264/H265 is awkward idea. Not sure what was the rationale behind. The H265 codec settings are producing extremely “fat” files for me.
Although version 3 looks better, GAIA processing is faster, there is support of parallel processing and multiple filters can be applied at single pass, the audio discarding and weird setting for bitrate are forcing to post-process TVAI 3.0.0 output files, which is waste of time. Previously output files were good enough to be used as final versions.
Hope this is something that can be fixed soon. Looks like minor issues, but I’ll stay on 2.6.4 for now
Thanks!
The VEAI uninstaller is not very elegant and seems to leave a ton of orphaned files/folders behind, especially when updating without a complete uninstall first. When the uninstaller failed to remove everything, I ended up manually deleting folders and then used CCleaner for the rest. The new V3 installer broke naming convention so it wasn’t grouped with other Topaz products.
There are 2 (maybe 4) environment variables that need to be set when running ffmpeg in a shell.
You can find them if you open the shell from Video AI. In the window, set | clip to copy all of the environment variables to the clipboard.
I’m certain on these two:
VEAI_MODEL_DATA_DIR
VEAI_MODEL_DIR
These are also present:
SESSIONNAME
ZES_ENABLE_SYSMAN
I’m writing a batch file to run my Video AI jobs that will have a notification when they’re done - over growl and rocketchat.
It should be including all tracks especially if audio is set to “Auto”. I suspect though the ffmpeg parameters are only specifying one audio track instead of to just copy all.
(The ffmpeg command string Video AI 3.0.0 builds has a number of issues.)
UPDATE:
Topaz Video AI 3.0 does indeed specify “-c:a copy” (well "-c:a" "copy" which is weird), it is not specifying -map 0:a, I’m not an FFMPEG expert but I think it needs that. (.e.g -map 0:a -c:a copy)
5 Likes
I’m on Windows and I only had to set VEAI_MODEL_DIR as per the documentation. I do have VEAI_MODEL_DATA_DIR set, but I think that’s leftover from the early access releases.
I have a python script that has grown into a monster, but it runs all the passes I want of TVAI and encodes the results into something more stream-able.
Thanks for the update and thanks for supporting non-square pixel sources. One note: If you choose to use a preset to upscale from non-square to HD or UHD, the software should automatically assume you want Square Pixel output. Currently, the software defaults to Original Pixel Type, which would give you wonky results.
What’s MKVToolnix GUI? When it comes to video processing, I’m a complete newbie. I’m just upscaling a bunch of lectures and training videos from the last 30+ years, and excerpting just the best scenes. Is MKVToolnix GUI part of Topaz?
No. It´s not related to Topaz in any way. It’s a very known software for remuxing video, audio and subtitles into a mkv container.
You can use google. It’s a good friend.
1 Like
I like the direction this is going! One workflow issue: I have jobs that are taking dozens of hours. I need to be able to pause & restart them so they don’t completely overwhelm my computer.
Under 2.6.4 I would just stop processing, set the start trim point to wherever the cursor started, then restart processing later. I would then join the separate segments into one output file.
I’m not at all sure how to do this with Topaz AI 3, but it’s unfortunately a deal breaker for me–I can’t afford to tie up my computer for 2 days at a time without being able to pause for video meetings, etc.
How can I get the same effect in Topaz Video AI 3?
