Trying to use topaz video ai (latest) from python... error loading models

Topaz Video AI Bugs and Issues
1

Hi

I’m trying to use topaz video ai (latest 3.1.6) on windows 11 x64 with an nvidia gpu, from a python script… if I run the ffmpeg.exe command from a cmd with the paths set, it works fine, but if I run it from python using subprocess.run (also tried subprocess.Popen to no avail), it always comes up with an error loading models.

the code: import osfrom pathlib import Pathimport subprocessVEAI_FFMPEG_PATH = ' - Pastebin.com
output of script running: > python main.pyInput #0, mov,mp4,m4a,3gp,3g2,mj2, from '70726847363002729 - Pastebin.com

DxDiag.txt (100.4 KB)

It’s very unusual, I’m not sure if ffmpeg.exe (your executable) is requiring additional things usually in a shell environment (?) or if it’s reading the path incorrectly? It’s reading auth.tpz correctly though.

I’ve noticed a slight difference with mention of “openvino” when run from python script & “tensorrt” when run manually from cmd (which works), perhaps I need to specify something else?

any help is appreciated
thanks!

2

I’m on a very similar system.
I use subprocess.call(command) where command is a string with everything I would put in the command line. Works correctly for me.
Because I need another version of FFMPEG to convert to H265, I do not set PATH. I always use the full path to the version of FFMPEG I need in each command.

Have you set the system environment variables for TVAI?

3

Yes I’ve set both TVAI_MODEL_DATA_DIR & TVAI_MODEL_DIR

4

Here’s an example of a command string I use:
"G:\Program Files\Topaz Labs LLC\Topaz Video AI\ffmpeg.exe" -hide_banner -stats_period 2.0 -nostdin -y -i "E:\Dump\somedvd\B2_t00 pt2 ffvpy.mkv" -sws_flags spline+accurate_rnd+full_chroma_int -filter_complex scale=w=853:h=480,setsar=1,tvai_up=model=prob-3:scale=2.25:w=1920:h=1080:preblur=0:noise=0:details=0.05:halo=0.01:blur=0:compression=0.6:device=0:vram=1:instances=1,scale=w=1920:h=1080:flags=lanczos:threads=0:force_original_aspect_ratio=decrease,pad=1920:1080:-1:-1:color=black -c:v png -pix_fmt rgb24 "E:\Dump\somedvd\B2_t00 pt2 ffvpyprot\%6d.png"
That’s the non-escaped string that gets printed out with the print function.
In the code there’s a lot of parsing and such:

command = TVAI + ' -hide_banner -stats_period 2.0 -nostdin -y -i "' \
        + os.path.join(filepath['-path'], filename + fext) + '"' + filt + ' -filter_complex scale=w=' + \
        filepath['-w'] + ':h=' + filepath['-h'] + ',setsar=1,tvai_up=model=prob-3:scale=2.25:w=1920:h=1080:prebl' \
        'ur=' + protpreblur + ':noise=' + protnoise + ':details=' + protdetails + ':halo=' + prothalo + ':blur=' + \
        protblur + ':compression=' + protcompression + ':device=0:vram=1:instances=1,scale=' \
        'w=1920:h=1080:flags=lanczos:threads=0:force_original_aspect_ratio=decrease,pad=1920:1080:-1:-1:color' \
        '=black -c:v png -pix_fmt rgb24 "' + os.path.join(outpath[0], outpath[1]) + '\\%6d.png"'

Try out a variation of that on your machine. If you still get the same issue, then I have no idea what’s going on.

My python script has multiprocessing setup so that I can have as many instances of TVAI running as I want. I did try piping the output from vapoursynth directly into TVAI with Python, and I never got it working. Seems it cannot call a pipe from Python. Or, I didn’t want to take the time to figure out how to get it working.