Here is the first of many performance beta releases. Since these releases have fundamental changes to the processing pipelines please let us know if you face any issues with processing or resource utilization.
Patch for Nvidia users having issues with model loading, replace the videoai.dll file with the following file in the zip in the beta installation directory. This will allow you to run all the models. videoai.zip (202.3 KB)
Major Changes:
New processing pipeline for enhancement
Almost full device utilization on all hardware and up to 2X speed up
Filter names have been changed to tvai_up, tvai_fi etc.
Environment variable changed to TVAI_MODEL_DIR for the json file directory and TVAI_MODEL_DATA_DIR for the model tz directory.
Known Issues:
Frame Interpolation, Stabilization have been disabled
NOTE: Even though it is labeled a beta many features are missing and will be added back in future betas. The app is labeled beta to allow automatic updates.
I just wanted to check with this. Have the AI models been changed in anyway to improve performance, or does the performance boost come entirely from the âprocessing pipelineâ changes?
Are there expected to be any negative side effects of the pipeline changes? E.G. Increased RAM/VRAM usuage?
Just installed, program says âloading modelâ, then goes to process but does not encode. RTX2070, windows 11, all previous versions of 3.x were fine. I will restart and check it again.
I appear to be able to make anything work as long as I donât upscale. For any upscale (tried 2x mostly) it downloads models loads models, then does nothing.
2 quick tests on a 1050ti Gtx
720 â 1080p 29.97
Proteus Auto : 0.82 spf (1.33 before !!! wow !)
Proteus Manual (Estimate) : 0.81 spf (around 1.21 before)
Beta not working on the A4500 RTX. it hangs on loading models. trying with lowering ram, and it refuse to redo a preview.
but seems to work ok on i7-7700/ 1050ti GTX card, on my side, but fail on AMD Epyc / A4500 RTX.
The speed seems to have improved by about 20%, however I canât get multiple clips to process simultaneously. The cryptic error message is:
Last FFmpeg messages:
Could not find codec parameters for stream 2 (Attachment: none): unknown codec
Could not find codec parameters for stream 3 (Attachment: none): unknown codec
Could not find codec parameters for stream 4 (Attachment: none): unknown codec
Input #0, matroska,webm, from âS:/Test.mkvâ:
track 1: codec frame size is not set
Output #0, mov, to âS:/Test_898161907.movâ:
The invididual clip processes fine as a single stream, but if a second or third one are added, it crashes.
Proteus auto 2x upscale, Ryzen 5950x, AMD 6900XT, 32GB RAM, Windows 10.
Also, the previous bug where a longer preview renders into RAM until all available RAM is consumed and then crashes once itâs out of RAM still persists.
Proteus and Artemis just waits at âloading modelâ. Gaia and Theia work as expected. I will try my other computers but I must stop the processing on them.
I also did an uninstall and deleted all Topaz beta directories before reinstalling, no difference.
Also, I have just noticed it seems to ignore the output file path set in âexport asâ. The output files are generated in the directory of the source file.
speed is stupidly awful. 3.5 minutes to render a 15 seconds preview of a progressive video from 720 to 1080 p @ 23 fps. with minimal enhacements from proteus fine tune/enhace. leaving a dxdiag with my hardware stats and a pic of the stupidly slow preview. not even attemping to export. and i choosed as coded the 265 AMD enhaced (i have an AMD gpu)
Played around for a littlebit. So far i notice the processing is slower and previews look very pixelated.
3.0.11 (17 seconds)
3.1b (23 seconds)
The spf is simular, but i notice the 3.1b gives several failed model warnings before starting.
On the Alpha post it said models are slow, but it doesnt say that here on the beta. (Proteus auto 1920x1080p)
(Looks a little worse irl, dont think i captured it at a high enough res. But as you can see, its pixelated)
Left: 3.1b
Right: 3.0.11
Update: After doing a short render the speeds where in fact better than 3.0.11, the âfailed model warningâ was to blame for the slower previews. Also, the rendered video did not suffer the pixelation the preview did. Though, it is hard to work on a video if the preview differ from the end-result
When I click onto Preview or onto Export or onto Export As ⊠nothing happens. This might related to the change of the framerate maybe. I also begins to save to the original file location instead of the one set via Export As ⊠Still unknown if it will move the file to the desired position after finishing. Anyway, it hangs on loading the model forever (?).
Things seem to be working on my RX 6600 machine. Only seeing about a 25 percent increase in speed with Proteus 540p to 1080p. Now to find out why my RTX 3060 and 3060 Ti machines arenât working. Might be a driver issue???
This might related to the change of the framerate maybe. I also begins to save to the original file location instead of the one set via Export As ⊠Still unknown if it will move the file to the desired position after finishing. Anyway, it hangs on loading the model forever (?).