You said: “go in your topaz account, there is a support page in the menu to open a support ticket !”
This option is not available AFAICS. I had to asked in the chat bot “How to report a bug?”. Topaz replied “To submit a bug report, reply to this email”.
I can not set the default preview zoom level to 50 % in the preferences.
Thank you very much for quick fix. I can confirm, that with my GTX 1650 slowness of previous versions is away and seems to be even faster than previous fastest versions.
You are right: Only those models that have changed should be downloaded.
I’ve checked bit by bit (program used: beyond compare) all models in model folders of v3.1.2 against v3.1.2.0b and they are all identical bit by bit (only the dates of files are different).
1 Like
“Add Noise” is applied pre-process, but “Grain” should be post-process and not affected by the model. Maybe its being affected by export compression.
1 Like
3.1.2 I am getting about 5.5fps on my 3080 Upscaling a 1080 to 4k + proteus with reduce noise and sharpening/deblur.
Yes. Fine grains are “wiped away” by the Nvidia hardware encoder. Even at high bitrate. Unfortunately. For this reason for my videos where I prefer fine grain the only way to go is to export as ProRes and then do software encoding. X264 can do that well. With X265 there are problems even with software encoding. These can perhaps be fixed with fine tuning. But I have not found anything yet.
I noticed a detail in 3.1.1 and 3.1.2 when I export in image sequence. It starts at frame 000001 instead of 000000 like version 2.6.4 did?
Moreover to avoid possible shift of sound, I duplicate the frame 000001 and I rename it to 000000.
All the same, the positive point is the speed of rendering.
I’m at 0.13-0.14 seconds per frame with an RTX 3070 GPU on a 1440x1080 to 1440x1080 video with artemis medium quality v13 to denoise and increase sharpness.
On 2.6.4, I’m at 0.24-0.25 for the same video. So in one day, I can make at least 3 videos of 49 minutes in only +/- 2h30 per video instead of more than 4h00 on the 2.6.4
2 Likes
Just double checked and its 100% getting affected by the model processing I changed my output to proresHQ and it still happens to the noise thats applied. I set grain and it doesnt seem to be affected.
Im finding with all encoders that Noise still gets “denoised” after the model processes. I don’t understand the point of adding noise if its going to get denoised by the model? This can’t be by design.
“Add Noise” is supposed to be affected by the model processing according to what the devs have said previously. I think it’s intended to help prevent over-processing and reduce artifacting, but I’ve never really tested it. It’s not intended to produce noise that you want to see in the final output (that’s what “Grain” is for).
2 Likes
With my AMD Ryzen 7 5800X @4,35GHz + RTX3060 I get 3x3.3 fps =9.9 fps (3 Videos in parallel) with full load on CPU (98%) and GPU (90-100%).
1 Like
Add noise before processing by the models is intended to help the model to produce better results in some types of scene, and is correctly denoised before rendering.
Add grain comes after processing by the models and is intended to make the rendered output look more natural or ‘filmy’, in cases where you want that.
But very high levels of compression during rendering can sometimes remove some of the grain you just added, in which case, compress it less and if TVAI doesn’t allow enough control over that, use Prores (a completely lossless option would be even better, come on Topaz!) then compress it to h265 or whatever in a program that gives you control.
Thanks for clarifying this!
I got 3.8 fps - 1m.10s
Also testet at the same video and settings.
Im running a 13900K, DDR5 Ram and RTX4080
1 Like
Were you using the CPU at the same time as the GPU?
In CPU mode, I am at 5.84 seconds per frame. I have an Intel I9 9900K CPU
7900 xtx gpu from amd here, still no av1 support just a heads up
1 Like
In preferences I’ve set “NVIDIA 3060”.
Windows Task Manager shows in “Processes” 3x ffmpeg.exe @31% and 1x Topaz Video AI @5%, while GPU is @93%.
