Topaz Video AI Beta 3.2.2.0.b

nipun.nath · April 13, 2023, 10:17pm

Hello everyone!

We have another beta release for y’all’s testing.

Downloads
Windows
Mac (DMG) | Mac (PKG)

Changelog:

[ Experimental ] A faster frame interpolation model (Apollo Fast) - only works in 4x slow-mo
[ Experimental ] Newer checkpoint of Proteus V4 for faster and slightly sharper results
Max processes limit is based on user hardware
Fixes issue with too much noise added the first few frames
GUI can display input/output up to 32000x32000 as long as there is enough RAM
Make patches target multiple baseline versions (smaller update installers)
Login and update workflow improvements
Disable audio for previews for better accuracy
Fixes preview mismatch for image sequences when framerate is not 25fps
GUI uses point size for all text, should work better with OS accessibility settings and large monitors
Preset indicates changes when filters are changed after applying a preset
Source video listing panel horizontally resizable
Misc updates

Known Issues

Split view doesn’t work correctly for input/outputs over 8K
Proteus V4 does only 4x upscale. So performance will be slow for lower upscales
Apollo Fast only works at 4x slow motion

Thanks for testing

Please upload problem videos and logs here: Submit files

david.123 · April 13, 2023, 11:22pm

Hello and thanks for another Alpha from y’all !

alanhusband-677420 · April 14, 2023, 5:48am

This new forced limit for “max processes” might be causing processes to fail because of hitting vram limits that weren’t being hit previously. I’ve noticed that “instances” has been changed to 1 from 0 (see benchmarks) due to “max processes” now being limited to 1 on my gtx1060. But from my understanding this internally turns off what used to be the “low power mode” option and hence allows 2 “threads” to run on that single process, increasing the vram requirement, which the gtx1060 can’t handle on some model/upscaling combinations.

Topaz Video AI Beta v3.2.2.0.b
System Information
OS: Windows v10.2009
CPU: AMD Ryzen 7 2700X Eight-Core Processor 15.939 GB
GPU: NVIDIA GeForce GTX 1060 6GB 5.9014 GB
Processing Settings: device: 0 vram: 1 instances: 1
Input Resolution: 1280x720
Benchmark Results
Artemis 1X: 3.87 fps 2X: 2.69 fps 4X: ERR fps
Proteus 1X: 3.79 fps 2X: 2.64 fps 4X: ERR fps
Gaia 1X: ERR fps 2X: ERR fps 4X: ERR fps
4X Slowmo Apollo: 4.35 fps APFast: 18.06 fps Chronos: 2.35 fps CHFast: 3.80 fps

Topaz Video AI v3.2.0
System Information
OS: Windows v10.2009
CPU: AMD Ryzen 7 2700X Eight-Core Processor 15.939 GB
GPU: NVIDIA GeForce GTX 1060 6GB 5.9014 GB
Processing Settings: device: 0 vram: 1 instances: 0
Input Resolution: 1280x720
Benchmark Results
Artemis 1X: 3.76 fps 2X: 2.57 fps 4X: 1.03 fps
Proteus 1X: 3.63 fps 2X: 2.50 fps 4X: 1.02 fps
Gaia 1X: 1.35 fps 2X: 0.91 fps 4X: 0.65 fps
4X Slowmo Apollo: 4.20 fps Chronos: 2.29 fps Chronos Fast: 3.70 fps

EDIT:

I was only able to get Gaia2X and Gaia4X working by making the vram setting very low, which has a significant performance penalty:

Topaz Video AI Beta v3.2.2.0.b
System Information
OS: Windows v10.2009
CPU: AMD Ryzen 7 2700X Eight-Core Processor 15.939 GB
GPU: NVIDIA GeForce GTX 1060 6GB 5.9014 GB
Processing Settings: device: 0 vram: 0.5 instances: 1
Input Resolution: 1280x720
Benchmark Results
Artemis 1X: 3.65 fps 2X: 2.54 fps 4X: 1.05 fps
Proteus 1X: 3.59 fps 2X: 2.51 fps 4X: 1.04 fps
Gaia 1X: 1.31 fps 2X: ERR fps 4X: NaN fps
4X Slowmo Apollo: 4.36 fps APFast: 17.95 fps Chronos: 2.20 fps CHFast: 3.64 fps

Topaz Video AI Beta v3.2.2.0.b
System Information
OS: Windows v10.2009
CPU: AMD Ryzen 7 2700X Eight-Core Processor 15.939 GB
GPU: NVIDIA GeForce GTX 1060 6GB 5.9014 GB
Processing Settings: device: 0 vram: 0.1 instances: 1
Input Resolution: 1280x720
Benchmark Results
Artemis 1X: 2.14 fps 2X: 1.50 fps 4X: 0.63 fps
Proteus 1X: 2.11 fps 2X: 1.48 fps 4X: 0.63 fps
Gaia 1X: 0.80 fps 2X: 0.54 fps 4X: 0.39 fps
4X Slowmo Apollo: … fps APFast: - fps Chronos: - fps CHFast: - fps

This all leads to the unintended side-effect that force limiting “max processes” can increase the gpu requirements on lower end gpus. Having a way to turn “instances” back to 0 would be useful.

20rushtonj · April 14, 2023, 10:40am

Out of curiosity, how is this determined? VRAM? RAM? Maximum number of concurrent hardware encoding sessions? Something else?

Because my hardware can definitely handle 4-5 concurrent jobs (typically one AI filter at 1080p or upscaling to 4k), yet it is now limited to 3.

Hardware:
GPU: RTX 4090 (24GB of VRAM)
GPU Driver: 531.41 (This allows for 5 concurrent hardware video encodes)
RAM: 64GB
OS: Windows 11

Rayburner · April 14, 2023, 11:07am

this beta still wouldn’t work with the Apollo model.

Proteus V4 on the other hand now works on this machine.

Testing now…

Topaz Video AI Beta  v3.2.2.0.b
System Information
OS: Windows v11.2009
CPU: AMD Ryzen Threadripper 3970X 32-Core Processor   127.9 GB
GPU: NVIDIA GeForce RTX 2080 SUPER  7.8135 GB
Processing Settings: device: 0 vram: 1 instances: 1
Input Resolution: 1920x1080
Benchmark Results
Artemis		1X: 13.73 fps 	2X: 10.36 fps 	4X: 3.07 fps 	
Proteus		1X: 13.53 fps 	2X: 10.17 fps 	4X: 3.03 fps 	
Gaia		1X: 4.25 fps 	2X: 2.88 fps 	4X: 2.04 fps 	
4X Slowmo		Apollo: ERR fps 	APFast: 10.73 fps 	Chronos: 5.66 fps 	CHFast: 5.98 fps

Imo · April 14, 2023, 11:24am

I am not sure if the app takes into account the Windows scaling
Each preview is put at the end of the list instead on the top
I suggest to get rid of the “Export” button and only offering “Export As”
Making changes to the Preferences resets the whole application
The Preview gets stuck still not following the processing
When used with frame interpolation (apollo fast) an unknown error ends any attempt to start processing

TPX · April 14, 2023, 1:08pm

@nipun.nath

Proteus V4 has problems with glasses, they did get added to the faces.

Then has something of a scary movie.

And the faces usually do not match what they are supposed to replace. (see text below)

Apart from that, it is a masterpiece what you have already made out of it.

I can understand when I see the input, what the model makes out of it.

Nativ res for improovement works best.

If TVAI was a main app I would have a 4090 by now, with AMD it’s a pain.

suraj · April 14, 2023, 3:13pm

Max processes is based on RAM & VRAM.
Mathematically it is min(round(log(2)(ram) - 0.2) - 2, round(log(2)(vram) - 0.2) - 1)

suraj · April 14, 2023, 3:13pm

Can you try deleting all tz files starting with apo in the models directory. Let us know if it doesn’t fix the problem.

Rayburner · April 14, 2023, 3:31pm

Hi suraj, there are no tz files starting with apo in the models directory (only apf, chf,…).

suraj · April 14, 2023, 3:46pm

Can you send me the logs. Are you able to run the Apollo model for preview or export?

Rayburner · April 14, 2023, 4:13pm

I am unable to run neither preview nor export. It sits for like 10 minutes Downloading and finally says AI error.
logsForSupport.tar.gz (350.1 KB)

So basically it looks like it can’t download the APO model

messages see below:

Error message from AI engine: download failed.
Error message from AI engine: model failed.
Error message from AI engine: download failed.
Error message from AI engine: download failed.
Error message from AI engine: model failed.
Error message from AI engine: download failed.
Error message from AI engine: download failed.
Error message from AI engine: model failed.
Error message from AI engine: download failed.
Error message from AI engine: download failed.
Error message from AI engine: model failed.
Error message from AI engine: download failed.
Error message from AI engine: download failed.
Error message from AI engine: model failed.
Error message from AI engine: download failed.
Error message from AI engine: download failed.
Error message from AI engine: model failed.
Error message from AI engine: download failed.
Error message from AI engine: download failed.
Error message from AI engine: model failed.
Error message from AI engine: download failed.
Error message from AI engine: download failed.
Error message from AI engine: model failed.
Error message from AI engine: download failed.
Error message from AI engine: download failed.
Error message from AI engine: model failed.
Error message from AI engine: download failed.
Error message from AI engine: download failed.
Error message from AI engine: model failed.
Error message from AI engine: download failed.
Error message from AI engine: download failed.
Error message from AI engine: model failed.
Error message from AI engine: download failed.
Error message from AI engine: download failed.
Error message from AI engine: model failed.
Error message from AI engine: download failed.
Error message from AI engine: download failed.
Error message from AI engine: model failed.
Error message from AI engine: download failed.
Error message from AI engine: download failed.
Error message from AI engine: model failed.
Error message from AI engine: download failed.
Error message from AI engine: download failed.
Error message from AI engine: model failed.
Error message from AI engine: download failed.
Error message from AI engine: download failed.
Error message from AI engine: model failed.
Error message from AI engine: download failed.
Error message from AI engine: download failed.
Error message from AI engine: model failed.
Error message from AI engine: download failed.
Error message from AI engine: download failed.
Error message from AI engine: model failed.
Error message from AI engine: download failed.
Error message from AI engine: download failed.
Error message from AI engine: model failed.
Error message from AI engine: download failed.
Error message from AI engine: download failed.
Error message from AI engine: model failed.
Error message from AI engine: download failed.
Error message from AI engine: download failed.
Error message from AI engine: model failed.
Error message from AI engine: download failed.
Error message from AI engine: download failed.
Error message from AI engine: model failed.
Error message from AI engine: download failed.
Error message from AI engine: download failed.
Error message from AI engine: model failed.
Error message from AI engine: download failed.
Error message from AI engine: download failed.
Error message from AI engine: model failed.
Error message from AI engine: download failed.
Error message from AI engine: download failed.
Error message from AI engine: model failed.
Error message from AI engine: download failed.
Error message from AI engine: download failed.
Error message from AI engine: model failed.
Error message from AI engine: download failed.
Error message from AI engine: download failed.
Error message from AI engine: model failed.
Error message from AI engine: download failed.
Error message from AI engine: download failed.
Error message from AI engine: download failed.
Error message from AI engine: download failed.
Error message from AI engine: model failed.
Error message from AI engine: download failed.
Error message from AI engine: download failed.
Error message from AI engine: model failed.
Error message from AI engine: download failed.
Error message from AI engine: download failed.
Error message from AI engine: model failed.
Error message from AI engine: download failed.
Error message from AI engine: download failed.
Error message from AI engine: model failed.
Error message from AI engine: download failed.
Error message from AI engine: download failed.
Error message from AI engine: model failed.
Error message from AI engine: download failed.
Error message from AI engine: download failed.
Error message from AI engine: model failed.
Error message from AI engine: download failed.
Error message from AI engine: download failed.
Error message from AI engine: model failed.
Error message from AI engine: download failed.
Error message from AI engine: download failed.
Error message from AI engine: download failed.
Error message from AI engine: .
Error message from AI engine: model failed.
Error message from AI engine: download failed.
Error message from AI engine: download failed.
Error message from AI engine: model failed.
Error message from AI engine: download failed.
Error message from AI engine: download failed.
Error message from AI engine: model failed.
Error message from AI engine: download failed.
Error message from AI engine: download failed.
Error message from AI engine: model failed.
Error message from AI engine: download failed.
Error message from AI engine: download failed.
Error message from AI engine: model failed.
Error message from AI engine: download failed.
Error message from AI engine: download failed.
Error message from AI engine: model failed.
Error message from AI engine: download failed.
Error message from AI engine: download failed.
Error message from AI engine: model failed.
Error message from AI engine: download failed.
Error message from AI engine: download failed.
Error message from AI engine: model failed.
Error message from AI engine: download failed.
Error message from AI engine: download failed.
Error message from AI engine: model failed.
Error message from AI engine: download failed.
Error message from AI engine: download failed.
Error message from AI engine: model failed.
Error message from AI engine: download failed.
Error message from AI engine: download failed.
Error message from AI engine: model failed.
Error message from AI engine: download failed.
Error message from AI engine: download failed.
Error message from AI engine: model failed.
Error message from AI engine: download failed.

suraj · April 14, 2023, 4:45pm

Can you uncheck Autodetect proxy settings in preferences → Application and try running Apollo again.

Rayburner · April 14, 2023, 5:28pm

still no change
logsForSupport.tar.gz (397.4 KB)

UHD4KUser · April 14, 2023, 10:00pm

V4 is faster now than in the previous release. Though I am still seeing this strange blotch issue as previously mentioned: Topaz Video AI Beta 3.2.1.0.b - #17

t9000 · April 15, 2023, 7:20am

Just had time to do a few small tests. Proteus v4 is giving me much better quality overall compared to v3. Very impressed so far. Can’t comment on speed, I don’t pay attention to this usually. Thanks for the update.

TPX · April 15, 2023, 1:30pm

Proteus V4 causes some errors.

My Cat is in the upload: Bf9i7877 (5 GB)

Was a testing video for working with a DJI Stabilizer.

Imo · April 15, 2023, 4:05pm

on videos of a smaller width I often can not set the preview zoom to the needed factor. while I am using a windows scaling of 200 % and video ai does not take this into account I always have to select 50 % of preview size manually. unfortunately when using the single preview view video ai refuses to accept the 50 % value. as a workaround I switch to the splitted view before/after where video ai accepts the 50 % value and then switch to the single view after where fortunately the value remains and the zoom level is perfect.

t9000 · April 15, 2023, 6:19pm

Experimented some more, the preview sync between before/after in split screen is closer than in some of the earlier releases, but experiencing another annoyance now, maybe it’s new, not sure – the preview position resets when the preview restarts. When I’m zoomed in on a part of the video at 100% (or any other zoom level except “fit”), the preview moves back to the center of the video frame every time it restarts. This makes it hard to look at a region of the video for very long in preview, I have to manually scrub back and forth so I don’t lose what I’m looking at.