Topaz Video AI Beta 3.1.0.4.b

suraj · January 16, 2023, 3:04am

Hi Everyone,

Here is the final beta for v3.1.0 before Tuesdays release. Please let us know if you find any issues. We finally managed to merge the installers for windows. Everyone on windows please do a clean install to get the best performance.

Downloads: Win | Mac | Mac Update

Anyone on Mac having issues with processing(crashes, speed etc.) please use this version and let us know if you are still having issues.
Mac | Mac Update

Changelog:

Windows now has single installer (please delete the models or just clean install)
Improved performance for Chronos, Chronos Fast & stabilization on Nvidia GPUs
Themis now uses GPU
Presets are fixed
Videos load correctly when default preset is set

Known Issues:

Apollo models have been disabled from the GUI

Thanks for testing

Please upload problem videos and logs here: Submit files

PS: In case anyone is wondering, TVAI Beta 3.1.0.3b was never released.

iMacmatician · January 16, 2023, 3:40am

Video AI crashes instantly when I drag a video file into the main window and when using the “Browse” button.

I installed the Mac version (full, not “update”).

suraj · January 16, 2023, 3:46am

Can you send me the logs.

bnumerick · January 16, 2023, 3:51am

I’m seeing some inconsistencies on the mac side in macOS 13.1. Open a 1080p/30fps video select upscale to 4k using Proteus then select preview and it hangs for quite awhile on my Mac Studio M1 Ultra 48 core GPU but eventually runs@5.7fps, I’m seeing .3fps on my 16" MBP M1 Max 32 core GPU even after fresh reboots, 3.1FPS on my 14" MBP M1 Pro 16 core GPU and 1.4FPS on my M1 Air 8 core GPU. I’ve had the issue on the Mac Studio and 16" MBP since the first beta. They both work with 2.6.4 and 3.0.12. The app seems to hang on all of the machines after a little bit too.

suraj · January 16, 2023, 4:14am

On first load the models are compiled and cached on Mac, this might take sometime. So if you did a fresh install or a model was downloaded this is expected. Send me the logs if the performance is still low on subsequent attempts.

phil.stopford · January 16, 2023, 4:14am

I wondered whether the Linux build will see an update in some fashion to align it with the public release code

robin.koehler · January 16, 2023, 4:21am

I can’t get any file to process on the 3090 system. I uninstalled using the system add/remove dialogue and the directory was completely empty when I installed the current beta.

No models being downloaded, just failing with the following message:

Last FFmpeg messages:
Setting ‘h’ to value ‘2160’
Setting ‘flags’ to value ‘lanczos’
Setting ‘threads’ to value ‘0’
w:3840 h:2160 flags:‘lanczos’ interl:0
Setting ‘out_color_matrix’ to value ‘bt709’
w:iw h:ih flags:‘’ interl:0
Setting ‘video_size’ to value ‘1920x1080’
Setting ‘pix_fmt’ to value ‘12’
Setting ‘time_base’ to value ‘1/30000’
Setting ‘pixel_aspect’ to value ‘1/1’
Setting ‘frame_rate’ to value ‘30000/1001’
w:1920 h:1080 pixfmt:yuvj420p tb:1/30000 fr:30000/1001 sar:1/1
Setting ‘pix_fmts’ to value ‘yuv422p10le’
w:iw h:ih flags:‘’ interl:0
auto-inserting filter ‘auto_scale_0’ between the filter ‘graph 0 input from stream 0:0’ and the filter ‘Parsed_tvai_up_0’
query_formats: 6 queried, 5 merged, 1 already done, 0 delayed
deprecated pixel format used, make sure you did set range correctly
deprecated pixel format used, make sure you did set range correctly
Output size set to: 3840 2160
Here Config props model with params: aup prob-3 2 0 1 1 1920 1080 0.000033 29.970030

2023-01-16-15-16-15-Main.tzlog (238.0 KB)

robin.koehler · January 16, 2023, 4:27am

Artemis, Gaia and Theia process the files fine, it’s only Proteus that aborts instantly and apparently even before attempting to download the model.

Same behaviour for preview as for export, no change when trying manual parameters, relative to auto, or auto.

Exporting the file without enhancing the resolution (1x) allows for the model to download, but the processing still aborts with the same message.

2023-01-16-15-25-0-Main.tzlog (824.4 KB)

robin.koehler · January 16, 2023, 4:42am

The 6900XT system does load Proteus and starts processing, however is still limited to only 2 parallel streams.

2023-01-16-15-39-14-Main.tzlog (494.5 KB)

thorr2 · January 16, 2023, 5:01am

I tried a TIF Image sequence with 1920x1044 source resolution @23.976 FPS. I selected the preset Upscale to 4K and convert to 60FPS. I changed the Frame Interpolation AI Model to Chronos. I left everything else as is. When I clicked Preview, I got

Error
Unknown Error

Last FFmpeg messages:

Unable to parse option value “0” as video rate

Stream mapping:

Unable to parse option value “0” as video rate

Error while filtering: Cannot allocate memory

bnumerick · January 16, 2023, 5:01am

Here’s the log file from my 16" MBP that is exhibiting slow render speeds.
2023-01-15-23-52-51-Main.tzlog (143.4 KB)

bnumerick · January 16, 2023, 5:08am

Here’s the log from my Mac Studio. Its also getting slow render speeds on/off. For this log it was only rendering .3fps compared to 5.7fps with the preview earlier (that too is slow now for some reason).
2023-01-15-23-56-2-Main.tzlog (74.4 KB)

kevin-1894 · January 16, 2023, 5:13am

Yep - * Videos load correctly when default preset is set - Fixed!
An export I ran last week on 3.1.0.0m took nearly three and a half hours, today with 3.1.0.4.b it would have been 1hr 15 minutes!

The only thing I noticed is that there seemed to be a lot more Halo on Proteus AUTO then previously? Will experiment a bit more but at this stage, 3.1.0.4.b is looking good!

i9 CPU - RTX3080 GPU - Windows 11

meimeiriver · January 16, 2023, 5:32am

@suraj I was going to get back on you, regarding the odd dithering I experienced with the beta. I took the liberty of testing it with this new beta.

At first glance, it looked like the issue persisted:

But I think here this was siimply the result of sharpening and extra ‘recover detail.’ Around the same frame:

I think we can safely say here the issue has been resolved.

A quick test on the old keypad:

No discernable distortion any more, beyond the same sharpening and extra ‘recover detail.’ So, whatever you did with those tensor cores, looks like you fixed it.

david.123 · January 16, 2023, 6:09am

Using Chronos Fast to convert a 4k 25fps source to 50fps (no other processing selected) no longer crashes like the last beta but performance is very slow (about 3x slower) compared to the 3.0.12 production release. GTX 1070.

20rushtonj · January 16, 2023, 6:43am

I noticed something. Changing the Max Memory Usage % value to something less than 100% results in a large increase in RAM/VRAM consumption.

This was noticed on a M1 Pro running MacOS.

Steps for me to reproduce:

Open TVAI GUI.
Import a video.
Change the scale to 2x and set the model to Proteus.
Start upscaling the video and take note of the RAM consumed by the ffmpeg process spawned by TVAI (Roughly 2GB in my case.)
Open TVAI settings and reduce the max memory usage to 90%.
Run the upscale again and take note of the RAM consumption. It’s much higher now (7GB in my case)

cdussud · January 16, 2023, 6:45am

When I 2x scale with Proteus and do 60fps with Chronos Fast, the video preview runs at half speed (slow motion). The exported file runs normally.

This does not repro if I keep either the original size or original frame rate.

Original file is 1280x720 at 30fps. M1 Macbook Pro

FlyingOfficer · January 16, 2023, 7:42am

@suraj
no apollo
Somehow I am not able to find Apollo Model under Frame Interpolation, is this expected or is it just me?

FlyingOfficer · January 16, 2023, 7:44am

My Apologies, just saw in the release notes that its disabled for this release