dawkinscm wrote:
brazen1 wrote:
dawkinscm wrote:

I hope so because while RIFE is still the smoothest the few remaining artifacts were horrible. In my testing I was able to remove every RIFE 25/26 artifact from all my test scenes using SVP. But if the new 26 heavy can remove those artifacts too then I'm there. I'm still impressed with what SVP can achieve though and if the new 26 heavy isn't that much better then I have no issue using SVP with LSFG smile

Unable to select 4.26 Heavy using Tensor RT.  It only appears and becomes selectable using ncnn/Vulkan and has OSD errors during playback.

You're doing something wrong. But it doesn't matter because it's not an upgrade. 26 Heavy has been out for around a year, this is just a fix for some conversion errors. It is better than before but still has artifacts.

I'm a fairly new user and trying to learn. Thank you for the reply. I use SVP for transcoding videos permanently I will revisit from time to time. I also use SVP processing in real time for watch and delete video. I've been using 4.25 and noticed the link for 4.26 heavy. Although 4.26 heavy is old, the site shows it was upgraded just a few days ago and thought I'd give it a whirl and see if transcoding times lessoned without compromising quality and how real time processing compared. Taking your word for it, I'll just avoid it. Thanks for the heads-up.

I'm hoping you or anyone can help me concerning model management. I'm using W11 25H2 with an RTX 5090 and SVP Pro 4.7.0.302 fwiw. In the past, using earlier builds, I d/l models. I unzip them to C:\Program Files\SVP 4\rife\models. I play video in real time and .engine and .engine.cache build and store at C:\Program Files\SVP 4\rife\models\rife. I have 27 different models all accessible and working using TensorRT. Being a newbie, it took me a while to narrow my models down to 4.25. Is that THE general consensus model I should be using? I'm not seasoned enough to figure it out myself.

So, I wanted to add 4.26 heavy and did it as the previous 27 models except it was not selectable and did not appear using TensorRT like the others. At this point I think I'm doing something wrong. Many uninstalls and reinstalls later, I'm getting nowhere however I did finally figure something out but I know it isn't correct: I open 4.26 heavy. There's rife and rife v2 folders just like every other model. Inside those are rife_v4.26_heavy.onnx files - again similar to other models. I copy the 'v2' heavy.onnx file to C:\Program Files\SVP 4\rife\models\rife. Still it isn't accessible in SVP control panel. In fact, I notice no 'v2' .onnx files from other models have ever appeared either such as 4.25 heavy and not just the 4.26 heavy I'm struggling with.

I finally replaced rife_v4.26_heavy.onnx file from the 'rife_v2' folder with the rife_v4.26_heavy.onnx file from the 'rife' folder instead to C:\Program Files\SVP 4\rife\models\rife.  Now, this model and others appear, are selectable, apply, and work as intended afaict.  This can't be normal is it? Are there left over SVP registry settings corrupting new SVP installs?  I'd search after uninstalling but no idea what to edit if anything. Are there other W11 SVP related folders/files not uninstalling? Etc.

This leads me to culling my models. I don't need or want all of them anymore and would like to learn how to manage them. Before I screw things up more than I assume they are, I thought I'd ask here because I'm not familiar or know if an SVP built-in tool exists to add/remove models? I'm tempted to delete the models I don't want from C:\Program Files\SVP 4\rife\models.  Next I would delete related .engines and .engines.caches from C:\Program Files\SVP 4\rife\models\rife. Any other suggestions?

On a side note, these are my advanced transcode settings I've been using. Log shows no errors but do they look ok in your opinion and would you suggest edits?  I know I am missing something because some metadata displayed info from the original file is no longer present after transcoding it when examining the files with the MediaInfo app. I've seen many files by others using SVP that did retain and display the info. No idea how they do it?

H.265/HEVC/NVENC/HDR:

b=120000000
preset=p7
rc=cbr_hq  (sometimes I use vbr also)
cq=14
qmin=15
maxrate=75000000 (I change this value depending on bitrate and file size desired)
minrate=70000000 (I change this value depending on bitrate and file size desired)
bufsize=120000000
tier=high
color_primaries=bt2020 (with Try to recover HDR colors unchecked)
color_trc=smpte2084     (with Try to recover HDR colors unchecked)
colorspace=bt2020nc     (with Try to recover HDR colors unchecked)
spatial_aq=1
aq-strength=12
temporal_aq=1
bf=3
g=300
refs=5
level=5.2 (using 4k chain now and will use 6.1 for 8k)

Hoping for any and all insight from the community which is always appreciated.

dawkinscm wrote:
flowreen91 wrote:
narkohol wrote:

I'm not sure about that. In all my test scenes RIFE is both smoother and has less artifacts than normal SVP:

I found this panning scene useful for smooth tests:

Slideshow Hero - S04E01 - 1080p
https://drive.google.com/file/d/1Ls-VdL … E_X7x/view

They just updated latest RIFE model to fix "abnormal artifacts" https://github.com/AmusementClub/vs-mlrt/issues/155 :
https://github.com/AmusementClub/vs-mlr … nal-models

Now we can finally upgrade the RIFE model after more than a year.

I hope so because while RIFE is still the smoothest the few remaining artifacts were horrible. In my testing I was able to remove every RIFE 25/26 artifact from all my test scenes using SVP. But if the new 26 heavy can remove those artifacts too then I'm there. I'm still impressed with what SVP can achieve though and if the new 26 heavy isn't that much better then I have no issue using SVP with LSFG smile

Unable to select 4.26 Heavy using Tensor RT.  It only appears and becomes selectable using ncnn/Vulkan and has OSD errors during playback.

Hi. New to SVP. Looking for general opinions from others about this configuration. Open to ANY suggestions. Video quality is more important than transcode time and file size but naturally less time and reduced file size would be welcomed without compromising visual quality.

Goal is high bit rate to enhance motion via interpolated extra frames. Most of my original files & Topaz outputs are in the 60 bit rate range. I'd like to retain near that after transcoding with SVP.

RTX 5090, i9-14900k & 64GB RAM.

TensorRT
4 GPU threads
4.25 (4.25 heavy FAILS)
H.265/HEVC/NVENC

opts:
b=60000000 (I assume this is another buffer setting) Yes/No?
preset=slow
rc=cbr_hq (using cbr instead of vbr to increase overall bit rate)
level=5.1 (will use 6.1 for 8k)
tier=high (helps honor minrate but falls below although higher than without. Also prevents FAILS using all of these settings)
qmin=15
cq=15
maxrate=63000000
minrate=58000000
bufsize=60000000
color_primaries=bt2020
color_trc=smpte2084
colorspace=bt2020nc

Questions:
(1) I'm transcoding UHD 23.976 fps to 60 fps aka 59.94x. Is it best to select x2.5 instead of 60fps?

(2) Using MediaInfo & other OSD's, all of my untouched & touched videos show video properties like HDR nits, Formats, Writing library, Mastering display info's, Light levels, etc. After transcoding using SVP, none of this info is present anymore. How can I force SVP to retain it? I have seen SVP outputs by others that indeed retain them. No idea how it was done.

Thanks for any suggestions re: my setup and/or questions.

4

(0 replies, posted in Using SVP)

Screenshot shows some media info that is missing after transcoding.  How can I preserve it?

5

(7 replies, posted in Using SVP)

Ahhhhhh.  Good catch.  So do you think that Topaz retains high bit rate because it uses 5.1@high?  Fwiw, it gives 2 choices - Main and Main10. I use Main10 codec and a 'Target bit rate' setting I put at 60.

Do you think changing the SVP advanced setting 'level=5.1' to 'level=5.1 high' would let me raise to 'minrate=55' and 'maxrate=63' for a new video or what would you suggest?  I'm going to try that right now.

So I've changed to 'level=5.1 high'. I raised the minrate/maxrate... for a new Topaz video I completed and put into SVP.  It's 7% into the transcode... AND IT HASN'T FAILED!!!!!  So happy you caught that.

I won't know the final output until it's done and that is a couple hours from now.  It looks promising.

Thank you, thank you so much for helping me.  Can't wait to see the magic SVP at a high bit rate does for this fast paced action movie.  I'll report back with the results.

6

(10 replies, posted in Using SVP)

Newest SVP build and nVidia GPU driver present avisynth and vaporsynth player problems.

I am using newest SVP 4.7.0.302 and 580.97 Studio driver. MPC-BE & MPC-HC crash opening any video file. If I turn off vaporsynth in players external filters, players open with audio but no video. If I also turn off avisynth, playback is fine rendering audio and video normally along with madVR, LAV, etc. Naturally, real-time SVP isn't possible.

If I uninstall and reinstall the previous Studio GPU driver, there are no problems except SVP now shows GPU acceleration isn't possible when I used to be able to select the RTX 5090 and Optical Flow.

If I use previous SVP build and previous GPU driver, there are no problems I'm aware of at all. Everything works as intended afaik. I'm surprised I've seen no other complaints of this. Then again, perhaps it's just on my end although I've no idea why or how since it's a fresh O/S install with my normal apps and adjustments as usual.  Now I see I'm not alone. You've asked the same thing I prepared this text for earlier.

Back at the time I tested using 4.6.0.294 which was good and 4.6.0.296 which was not good.  Uhg!  2hr time limit between posts.

7

(7 replies, posted in Using SVP)

When transcode fails, first thing I do is check the log.  No error shows other than something about "out of parameter (8)" if I recall correctly.
I feel SVP must somehow care because when I enter a higher value, it fails.  Enter the lower value, and it doesn't fail.
Yes, I too checked Wiki. One of the changes was to enter 6.1 but that made zero difference when I tried it.
Perhaps the Wiki is wrong as is Google Ai because you can clearly see from my Topaz output file screenshot (and the original disc file) they are level=5.1 with bit rates well above 40 - 63.1 to be exact.
Even the SVP output is higher than 40 - 42.6.
That's why I don't trust everything I read and show examples of real world results. This is why I am seeking help here. I know you are knowledgeable.
Thank you again for taking interest in helping me.

8

(7 replies, posted in Using SVP)

Yes, I agree but that's not the problem.
The problem is SVP will not accept higher values than 43 and 44 or it immediately fails.
The original file is bit rate 63.1 and I'd like to end up close to that.
However, no matter what I try, SVP will not produce anywhere near 63.1 for any of the files I've transcoded.
The best I can get it to produce is 42.6 using the advanced settings shown.
I would like to increase minrate=43 and maxrate=44 to something like minrate=60 and maxrate=63.
But once again, SVP transcode will fail every time if I increase either of those values.
I guess my question is... why is SVP transcode failing when I use minrate=45 and maxrate=46 for example?
Is there a better advanced code to try although I have tried quite a few others.
Thanks for your reply and help.

9

(7 replies, posted in Using SVP)

I'm trying to retain close to the original file size and\or bit rate of original file after transcoding with SVP.

I'm using Windows 11 24H2 - RTX 5090 Liquid Suprim - i9-14900k - 64GB DDR5 @6400. After selecting a file to transcode, I've been selecting Rife Ai engine from the dropdown instead of Automatic. I've set mpv64 and FFmpeg paths in environment variables fwiw.

1st screenshot is SVP advanced settings I'm currently using - these have not failed and work ok. I've tried many, many others including lowering values of the ones listed and even removing some of them and replacing with others and/or omitting them altogether. Transcodes either fail or do not achieve a higher bit rate nor file size. I've not been able to raise 'maxrate' or 'minrate' any higher.

2nd screenshot is RIFE options. I've also tried other models. I've lowered GPU threads down to 1 but the transcode was slow and GPU usage was only 25%. Normally it is 73% so I put it back to 4. I did read somewhere 5090's require 1 or it would fail. Fwiw, mine doesn't.

3rd screenshot Topaz file size (52.2GB) and bit rate (59.3)

4th screenshot SVP's end result file size (38.2GB) and bit rate (43.4) That's the reduction I haven't learned how to cure.

I've completed a bunch of transcodes in the short while I've been amateur learning SVP from square one. All of them turn out similar to this one and all of them I've tried different things. I'm exhausted. I notice most people complain about file sizes and bit rates too large. My query is just the opposite.

Thanks for any help pointing me in the right direction others might offer.