I’ve seen that too in order versions.
See if the same thing happens using different input/output formats.
Most of the stuff generated by VEAI is (intentionally) done by default in the temp file areas. As such, (in Windows at least,) opening a specific drive’s properties, and selecting “Disk Cleanup,” making sure that Temp files is selected and (obviously) Recycle bin, should get rid of most of it.
I also wonder what ffmpg does exactly when the gui crashed. Yes, ffmpg finishes the job and the result is looking good. So are the frame by frame auto adjusting mode is not continously done by the GUI so that ffmpg continues even this on its own when the gui got lost? 
You could also terminate the ffmpeg process from task manager.
All the processing is done by ffmpeg including any enhancement filters that you might have selected before the GUI crashed. ffmpeg doesn’t really need anything from the GUI as all the GUI does is generate and pass parameters to it. So if you selected Proteus Auto, ffmpeg will continue to process your source while applying that enhancement.
1 Like
I know that the processing is done by ffmpg. I wonder what does the individual frame adjustment guess for each frame to be handed over to the ffmpg processing when the gui is gone.
… and what the hell is VEAI doing (CPU Load: VEAI 12%, ffmpeg 5%)
2 Likes
There is a way to use Topaz Video AI 3 in VirtualDub2 as an external encoder (CLI), so you can load any avisynth script or use VirtualDub filters as pre-process
You will have to configure and create external encoders profiles. Example with Artemis V13 and H264 NVENC stream output with no sound.
Open Notepad and copy this text
{
"description": "VirtualDub external encoder profile collection",
"externalEncoders": {
"sets": {
"TOPAZ VIDEO AI (video)": {
"videoEncoder": "FFmpeg TVEAI",
"audioEncoder": "",
"multiplexer": "",
"description": "(MP4 - H264)",
"extension": "",
"processPartial": false,
"useOutputAsTemp": false
}
},
"profiles": {
"FFmpeg TVEAI": {
"name": "FFmpeg TVEAI",
"program": "C:\\Program Files\\Topaz Labs LLC\\Topaz Video AI BETA\\ffmpeg.exe",
"commandArguments": "-f rawvideo -s %(width)x%(height) -r %(fpsnum)/%(fpsden) -i - -filter_complex veai_up=model=alq-13:scale=2:device=0:vram=1:instances=1:threads=0 -c:v h264_nvenc -profile:v high -preset medium -pix_fmt yuv420p -b:v 180M \"%(tempvideofile)\"",
"outputFilename": "%(outputname).m4v",
"type": 0,
"pixelFormat": "yuv420p",
"inputFormat": 0,
"checkReturnCode": true,
"logStdout": true,
"logStderr": true,
"bypassCompression": false,
"predeleteOutputFile": false
}
}
}
}
Save this as TVAI.vdprof in the VirtualDub2 folder (or anywhere else, it doesn’t matter)
Open VirtualDub2, go to ‘options’ — ‘External Encoders’
A popup window will open, click on ‘import…’ and choose the vdprof file just created and voila ! Click OK
Then when saving a video : File > Export > Using External Encoder and choose TOPAZ VIDEO AI
You can just create several profiles with any model or output desired or just edit the command arguments directly when needed.
Of course you still will be limited with the defaults encoders available in V3 unless you have build your own FFmpeg with veai.
1 Like
salut, j’ai trouvé un petit bug d’artefact lorsque j’ai convertit ma vidéo, avec les versions beta par rapport à la version 2.6.4. Je ne sais pas si ça a été mentionné déjà mais en utilisant artemis aliasing or moire, le nom CEZANE du bateau est flouté sur quelques images successives :
1 Like
cpu 50 % gpu 40 % on my machine.
There is a new setting that may be confusing you.
You have to choose “relative to auto” or “manual” from the dropdown.
On relative to auto, 0 is auto and the sliders are a percent adjustment up or down.
On manual, 0 is 0 and nothing will be done with anything set to 0.
After doing a Proteus preview on Auto and then switching to Relative to Auto preview froze up, zero entries or not… It was necessary to exit VEAI, restart the app, then go directly to Relative to Auto in order to do a preview.
BTW: Considering the poor input I’m feeding Proteus, it is doing a remarkable job of enhanced resizing. There are still some problems, but the image in general is cleaner and clearer. 
mh yes. I don’t like the splitted menu. It just feels like a bad implementation as an excuse of not knowing how to explain easy how it is meant to be working
People couldn’t understand it the way it was and kept complaining. This is what happens when there are too many voices.
I can confirm that changing from a 2-step stabilize to a 3 step stabilize with Rolling Shutter Correction and Reduce Jittery Motions both on, Auto-crop mode, stops the minor pixel repetition artefacts at the edges of the frame. They were most notably on the RH edge of mine. The stabilizer does a good job on severely bad footage (12 year old consumer cam at max zoom hand held on the back of a jeep with people moving about). The only remaining minor is a slow wave from movement of the camera’s IBIS. But the effect is minor and slow, and the subject of the footage (a cheetah up a tree) is now well observable whereas with the original you had to “imagine” a lot of it.
Now, just to put the deinterlace back in its own section so the best deinterlace can be selected AND enhancements can be applied after stabilization.
Processing with these settings is mighty slow, but for small clips that need savage correction, it’s doable.
Minor cosmetic problem: can’t reproduce reliably, but at some stage while playing a Preview, the RH pixel column of the video gets played in the vertical divider bar between the video and the controls on the RHS.
Don’t understand: When you select 2-pass stabilize, the Preview calculation status shows “1/2”. When you set it to 3-pass, it says “1/2” not “1/3”. Not sure what this means.
I can confirm in 2009b that not only does the Export button do nothing, but choosing Export As… and entering a file name / OK also does nothing at all. There is no indication of any processing and no output file is either created or even locked for writing. Task Manager shows that ffmpeg has NOT been launched.
1 Like
seems this “issue” doesn’t affect everybody. 3.0.0.9.b is working here, export button work, export saved where it must be. Export As button is working as well. Intel H264 new codec export work too.
there is certainly a common point between you and the ones who have this issue. did you had the “model folder” issue like some other people ?
When you’re comparing Previews in Split mode (original on left, processed on right), when you click between one Preview and another, the split bar is reset to halfway across the screen. Also, sometimes but inconsistently, it repositions the video content centered in the playback frame. Can the split bar and frame position please stay fixed for each Preview?
Later feature request: (AFTER the app is made reliable) Implement a Preview LOCK button that sets all Previews to the magnification, frame position, and split bar you’re currently working with. That would make Enhancement parameter testing workflow much faster and easier, to achieve maximum enhancement with minimum artifacts.
1 Like
3009b still is unreliable re GUI behavior and disappearing. It opens full screen even though the full screen window option is not selected. Clicking full-screen button results in no change, other than the Maximize icon changes to an Unmaximize icon. Clicking that does nothing at all. It says full screen.
Coupled with the GUI randomly disappearing even though it’s still running, and the fact that it also totally HIDES Task Bar, it’s VERY badly behaved. The Task Bar is pushed off screen whether or not it’s auto-hide or permanent. Seriously bad app behavior.
While testing this functionality, which required minimizing the 3009b window so I could use the Task Bar, after several minimizes, maximize now doesn’t work at all. Like 3006b (the last beta I tested), VEIA beta is now invisible and cannot be seen on screen AT ALL by ANY MECHANISM. Since 3006b, even the last resort, Maximize off the app’s shortcut menu in Task Manager, is no longer available. For some inexplicable reason, another completely different shortcut menu is now shown and it doesn’t have Maximize on it. It’s now WORSE, not better than before.
Right-clicking the app icon in the Task Bar and choosing Close does nothing at all. This is more seriously bad app behavior. You have to go kill of the app in Task Manager.
Now on relaunch, the GUI is still completely invisible and unusable. So I’ve uninstalled it and I’m not going to do any more beta testing until you fix this disaster properly. Please make the very basics play nice before toying with more nice-to-haves.
Please stop wasting your beta testers’ time. I vote that this is your one and only TOP PRIORITY for the next beta. Get the fundamental GUI stable: if you want us to look at it, make it reliably look-able at-able.
I also recommend that you write a small “cleaning” app for beta testers and users that thoroughly, reliably, 100%, uninstalls the Beta (but doesn’t touch 2.6.4 which is still working great) re ALL files (including dumps, configs, etc), folders AND Registry, so that we have clean machines to test a Beta install on without tons of manual labor and iterative comms with the devs or support.
Please advise directly if/when you have a stable GUI. THEN I will return to beta testing.
I see attempts to make the UI better, but sometimes they fall rather flat. like they are in need a good UI person. For example, the timeiline trim mechanism in 2.6.4 is great. The 3.x beta mechanism is far less intuitive, it’s clunky, harder to use, less accurate, and more jerky.
Re the Enhancement adjustment types, the labels are incoherent… “Auto”, “Relative to Auto”, “Manual”. ANYTHING that’s not full auto is manual! That is, “Relative to Auto” is also “manual” because there are controls (not present in Auto) and you manually twiddle with them.
These would be more obvious labelled “Full auto”, “Auto adjust”, and “Fixed”. Because that’s what they do, especially via the values on the sliders that show when you choose the mode: -100/+100 with 0 in the middle for “Auto adjust”, and 0-100% for “Fixed”.
I would also argue that there’s plenty of UI room to present these three options as a single row of buttons, active one highlighted. That makes the entire MENTAL SCOPE of the three forms of control immediately obvious, and changing is ONE click. A dropdown is NOT natural UI furniture for changing the current display (show/hide sliders). Tabs or buttons are the natural UI furniture for changing UI subspace (ie what’s presented under it). Given one of the options (Full auto) doesn’t have subspace, I’d argue that a set of buttons would look better than a set of tabs, one of which would be empty.
With a dropdown, first you have to NOTICE it, because only one item is shown and it doesn’t stand out. THEN you have to DROP DOWN the menu to see what’s available. THEN you have to examine the options and form a mental picture of their relationship (because it’s a vertical list of text floating in space rather than a horizontal row of buttons naturally FRAMING or HEADING the slider space below). THEN, having formed a mental picture, you have to click the option you want, or to dismiss the dropdown.
So, three well-labelled, permanent buttons visually and mentally scopes the options, frames or heads the following content (if any), and allows option selection with one click. That’s it. Most of the hard work is done at the design stage rather than by 1000s of users at run time.
(And the “Fixed” option includes a button [ Suggest settings ] that parses a bit of the input clip and sets the sliders to some useful values.)
4 Likes
