The user was using TVAI, don,t know which version.
After 3.1.0 was released
I asked about CPUs used.
In my main job I always have to assert myself because no one believes me at the beginning until they see it for themselves.
The work that TVAI does here with the PAL videos on the GPUs is not much, which is why people on the forum don’t see any scaling from a 3060 to a 4080.
The GPU is so fast with the work done that it is hardly relevant at small resolutions what optimizations you bring in there.
Because you would have to quintuple the performance for a faster execution to notice a difference.
You can achieve that with the Hopper H100, but it costs 10,000€, and no user will buy it.
“Export Without Preview” is now by default. While processing now it shows permanently only “… Generating Preview”. Since that is useless it should be removed.
Thanks guy’s for this beta update.
I got only some issue’s now with export the .mov files. It looks that are now render with a other codec ore something. I can’t open it Adobe Premiere.
Also the render time is very long.
Sample clip 2 m 10 sec render with 22 minutes. Normally it was about couple minutes finishend.
1920x1080 25fps
Frame Interpolation Apollo
Enhancment progressive
AI Model Proteus
Parameters auto
My pc setup.
NVIDIA Geforce RTX 3060Ti
i7 @3.80GHz
Mem 32 GB DDR4 DRAM 1066.2MHz
When installing it says old models were find and I say yes to delete them but instead absolutely all the models have been deleted and now I have to redownload everything.
Please test the following items and share your feedback:
Please test Stabilization and let us know if it still shows artifacts where entire frame has some colored patterns
I am still in the middle of a big job, but what I noticed was that the horrid blurry artefacts are from the “reduce jittery motions” setting with the stabiliser model, I’ll send my logs once I’m finished!
In my opinion he said that, no doubt they won’t do it, otherwise he wouldn’t answer that, so I think the manager models we can forget him, finally no doubt it’s no longer news, without hope
To me, it read more like “We may eventually get to it, but don’t hold your breath waiting for it, as our plate is full, and this is a very low priority thing for us.”
Yeah, noticed that too. But “Export Without Preview” didn’t produce a readable file (for my LibavSMASHSource VS filter, only one that can read ProRes 422 HQ), so I’n basically back to square one, needing a 2nd pass.
In that regard, I still don’t get why they cannot simply use the ProRes writer from v2.6.4; it had Preview, produced a readable file for my VS filters, and was single pass.
Is it planned to give us control of the order of execution for the different processing methods? To ensure something like motion deblur maybe happens after enhancement for example in some cases. I assume the order of the processing is top down / the same as the order of the logos in the preview output area for each preview.
Also it would be nice to be able to select up to 4 previews of the same time length, and play them side by side with frame by frame control etc. so we can compare outputs against each other more easily instead of always against the original.
A very sensible question! Does TVAI denoise first? Sharpen first? Etc. Would really help our fine-tuning if we knew the order in which the ‘sliders’ (Proteus) are applied.
That’s what I was thinking too yesterday. Themis can introduce a visible checkered pattern that then can be exacerbate when upscaling and in some tests I made I had better results with first Upscaling then proceeding Motion Deblur. Same with Apollo sharpen, better results after upscaling.
Rendering times could be longer but after all, quality still prevails over speed. At least for me.
So yes this should be an option given to choose the orders of processes. I don’t know where this idea of upscaling as the last step being the normal workflow come. It’s the same with Photo AI, many images are destroyed when doing denoise/sharpen first then upscaled when in some tests I’ve made I had excellent results when doing first the upsacaling then Denoise or/and Sharpen.
Peculiar. The notion of doing pre-denoising (with your own, advanced denoise filters) makes a great deal of sense: the less crap you carry from the source video, the less TVAI will (not so intelligently) try and do something wrong with it.
Which really, again, boils down to the order in which TVAI decides to do things: maybe indeed denoises only as a last pass? Some sort of dev feedback on this would be highly appreciated.
FYI:
I run Proteus on a 30 fps 1080 clip but downscale to 540 and save as prores.
That is then run thro GMFSS-Union to 60fps.
And then thro Resolve with x2 Superscale, minimal Temporal and Spatial NR, colour correction and a small amount of sharpen.
Finally I run thro av1an with svt-av1.
This workflow works best for me.
Point:
VAI is a great tool if used correctly.
As a stand alone it’s NOT a one click 240 to 2160 magic tool as some expect.
Although, it doesn’t look like they have deblur on that page yet.
If you want confirmation on which filters are being applied in which order, you can open up the app, setup your processing, then select from the top of the app Process -> Show Export Command. I’ve found the order in which the filters are listed in the export commands is the order in which they occur.
And if you’re willing too, you can edit the order the filters are processed in the export command or apply multiple filters of the same type (something that isn’t supported in the GUI).
