Search options (Page 12 of 17)
flowreen91 wrote:My suggestion would be to delete cache folder from here:
"C:\Users\username\AppData\Roaming\SVP4\cache"
every time you update rife models and let them generate fresh.
This is not necessary. In the SVP rife folder, just replace the onnx file for the model you want updated then delete the generated model and engine files. This will trigger regeneration. But if you are using model names that are different to the actual model being used then I don't know since I stopped doing this some time ago.
flowreen91 wrote:A v2 jaggedness/zigzag bug was fixed on 3 November
It was never a v2 issue, it was a model issue. The only v2 issue that has been recently fixed was lower quality output from source material over 2k.
RickyAstle98 wrote:
They have pattern match and random zigzag artefacts, where too many same lines onscreen or objects, interpolated by error too, regardless of the overall frame!
Model 4.9 almost eliminates that effect, even with very low frame rate videos, but more GPU sensitive...
This is correct. But some users may not have seen the issue depending on the videos they are watching. A good example of this effect is the Blade Runner 2029 intro with the patterned landscape. The zigzag artefact can be seen clearly just before the ship flies overhead with v4.6. but is minimised and only occurs once the ship flies over in v4.9. In v4.12 the artefact is gone.
scb wrote:So what about 4.13v2 lite, how does that rank?
Don't worry about the Lite versions, at least not yet. If Rife 4.9 v1 or v2 works for you then stick with that. if it's too much then use Rife 4.6 v1 or v2. But if you have a 3090 (or maybe 3080) or above then you can also try 4.12/4.13.
RickyAstle98 wrote:What happens when someone upload onnx packages for 4070 users? Will work for anyone who has the same GPU or not? I have v1/v2 prebuilt for HD/FHD sources with TRT boost enabled!
What about reverse engineering method for 4060Ti and 4060 owners too?
Average onnx build time v1 ~1/1.5 min HD/FHD
Average onnx build time v2 ~2/2.5 min HD/FHD
There are no specific onnx builds for 4070 or any other users. What you have are builds that the 4070 has enough power to process. In general, higher build numbers tend to require more powerful GPUs. I think Rife 4.6 and Rife 4.9v2 are you best bets but other 4070 users might be more helpful here.
Chainik wrote:sounds like a bad advice that others will follow blindly and get a worse results (as usual when you're following some advices w/o understanding them)
Maybe I should have said that hwdec off works better "for me".
aloola wrote:In my case it's improvement.But in some cases, if the CPU is so weak, the performance would drop a lot.
Same here. My wild and totally uninformed guess might be that there's more GPU resources available to process RIFE when not using hw decoding. I did a quick check of CPU resources with hwdec on/off and there wasn't any obvious difference. For both hwdec on and off, a couple cores were close to max, a couple were about 50% and the rest were either idling or parked.
flowreen91 wrote:Is there any simple way to test the difference between any two configurations?
My test is visual. With hwdec off I get smoother playback. YMMV.
aloola wrote:Turn off GPU decoder, and only use software decoder in player
Interesting that you say that. I don't know if you mean with mpv player but for SVP this player does work better with hardware decoding off.
mengran123 wrote:Hello, I have already spent money to activate the software, but I cannot use RIFE AI engine to patch frames in the VR software HereSphere VR Video Player. Once I use RIFE AI engine, I will report an error----
please help me！ I want to use RIFE AI engine to patch frames in the VR software HereSphere VR Video Player, thank you！
I just tried Heresphere and I will probably never touch it again. But thanks for the heads up 
Xenocyde wrote:dawkinscm wrote:Xenocyde wrote:I did not see it try to cache again after, so don't think it crashed the first time.
It doesn't cache in 15 seconds. None of them do.
No even on an RTX 4080?
Nope. Not even on an RTX 4080.
Xenocyde wrote:dawkinscm wrote:Xenocyde wrote:Woah, the default 4.9 model you added cached in like 15 seconds. The 4.9 I added some time ago took a minute and a few seconds to cache.
Then it's probably crashing before it completes.
I did not see it try to cache again after, so don't think it crashed the first time.
It doesn't cache in 15 seconds. None of them do.
Xenocyde wrote:Woah, the default 4.9 model you added cached in like 15 seconds. The 4.9 I added some time ago took a minute and a few seconds to cache.
Then it's probably crashing before it completes.
Ironclad wrote:Reading through the thread it looks like the current consensus for worthwhile models from increasing quality/decreasing performance are 4.4v2, 4.6v2, 4.9v2/4.12_litev2/4.13_litev2, and placebo tier would be something like 4.13_ensemblev2 I guess? 4.12/13 are apparently better for animation, but otherwise slower than 4.9.
Everything from 4.9 onwards was for anime, but they also work very well with live stuff. Rife 4.12/4.13 are better for fast motion which is why I now use them, but I'm watching on a very large screen. Rife 4.9 is the best quality/performance compromise for most people, followed by 4.6. I agree with your placebo tier.
cemaydnlar wrote:redraiderj wrote:cemaydnlar wrote:I tried both disabled and enabled performance boost for v2 but still not working. I know that normally even a 3090 would struggle but i am watching animes on a 1080p monitor and use downscaling method from svp than i upscale it back to 1080p with anime4k. I use x3.
You have to put v2 models under rife_v2 folder, not the rife one, the full path would be like C:\Program Files (x86)\SVP 4\rife\models\rife_v2
Now its working thank you.
A few posts back I did tell you to delete the v2 folder and re-download them. You didn't put the v2 files in the v2 folder?
cemaydnlar wrote:Tried everything seems like v2 does not work on 3070 plus i am having frame drops on 4.13.
A 3090 would struggle with 4.12+ that is why 4.9 is the default.
cemaydnlar wrote:They should somehow start to find a solution to make it use less gpu.
They can't at the moment. The models are getting better which is why 4.9 is the default, not 4.6. But as they get better they become more compute intensive. I assume that it would take a paradigm shift in model building before that changes. I'm just hoping that it gets just better enough to remove any obvious remaining artefacts without requiring a 4090.
cemaydnlar wrote:I can't use v2 on any version of rife. Do you guys know why plus %100 sct makes soap opera effect dissapear so i switched to %20.
You can either completely uninstall SVP then reinstall. That should fix everything. Or you can delete the files from the v2 folder and download the models again since there was a mixup with some of the model files. If you don't know where the files are, then reinstall and follow the instructions in the RIFE AI FAQ. That way you can be sure everything is setup correctly.
After a bit more experimentation I've found that Rife v4.12 with SC=7.15 works best for my VR setup so far. It is the best balance I've found between fast movement and slow panning. Reducing SC like this make the fast action a little less smooth with occasional repeated frames, but still clearer overall because the artefacts are gone. With the most recent builds of mpv I am able to match my VR fps using Fixed 60fps so maybe that helps. It seems to reduce GPU usage as well which I assume is because it's not interpolating as many frames.
Rife v4.13 is out. No obvious difference with v4.12 for me, maybe others will find something. They've also re-released Rife v4.7-4.12. No major differences found with v4.9.
Interesting. I found 9.1 to use a bit more of everything with no obvious improvements and the Dev notes seem to confirm my findings.. I think 9.2 just a fix for the crashing issues but let us know what you find.
Gippy wrote:Has RIFE become any faster? Do you still need a 4090 to go from 4k24 to 4k120 in realtime, without the use of TensorRT pre-caching? Or is a less powerful card good enough now? I'm considering a 4070Ti but am concerned that just won't be fast enough for RIFE 4k24 to 4k120.
No it's not becoming faster. After v4.6, Rife is effectively becoming slower because the models are getting bigger and more compute intensive. Rife v4.6 is about 20% faster than v4.9 which is about 20% faster than v4.12.
redraiderj wrote:reconhell wrote:When you download new rife you get 2 folders. normal and V2, which to use?
Also which is better lite or normal rife?
I'm using 3070, v2 doesn't work at all for me, while normal one gives me a lot of stutter, I have to switch back to 4.9. I think you can just try both and see how they work.
For live video the Rife 4.12 handles some fast motion better but for the most part you won't notice a difference compared to Rife 4.9. So depending on what you are doing, most users shouldn't bother with Rife 4.12. I know with my workload even a 3090ti would struggle.
Xenocyde wrote:flowreen91 wrote:P.S. OMG setting "Scene change threshold" to 100% makes the video look amazing just like the old "Uniform (max fluidity)" setting from https://www.svp-team.com/wiki/Manual:FRC
Why did nobody mention this before? Will only use this from now on. Many thanks!
Woah, I switched SCT to 100% and ya, can't really see any artifacts and sudden camera moves seem almost perfectly smooth on default 4.9.
I checked with my test files and anything over 15 gives a similar or the same effect for me which is basically what I have seen before. The major fast motion artefacts are still there, but this time I did see an improvement to a minor artefact. So for now I've set mine to 20 and will see how it goes. It would be interesting to see what the Devs think about setting to a 100.
Blackfyre wrote:@Chainik regarding "Scene Change Threshold", it's interesting how it works like artifact masking in the past. 15% provides the smoothest panning shots and the most artifacts. 6% shows the least artifacts, but less smoothness. 1% (set manually from all settings) has ZERO artifacts, but LEAST smoothness.
I use the same 3-8 movies scenes to check for panning smoothness and artefacts. I've tried SCT set to anywhere between 1 and 20% and 12% is the best balance I've found between artefacts and smoothness, especially with Rife 4.9 and above. I think I will play around between 13 and 15% for future testing.
reconhell wrote:How do I update to the latest version of RIFE for live conversion?
...
Why does my SVP screen does not show "engine" or "performance boost"?
Also I tried this and i have the files on location like in screenshots. but the dropdown box in SVP doesnt show any new option...
- Uninstall SVP completely.
- Reboot your machine
- Follow the instructions on this page https://www.svp-team.com/wiki/RIFE_AI_interpolation.
- Choose Rife 4.9 or if your GPU has problems then choose Rife 4.6.
Harze2k wrote:Just one more question you might know, when using svp with plext htpc. i cant see the ffmpeg.exe in any folder on my device, is that somehow included somewhere else?
a lot of the logs in mpv that points to issues, fails or error has ffmpeg in them but i cant find that executable anywhere.
MPV uses ffmpeg libraries for codecs, shaders etc. So even the Plex "fixes" are fixes to ffmpeg not mpv. The rate of development for ffmpeg is ridiculous considering the size of the project: https://github.com/FFmpeg/FFmpeg/tree/master
Harze2k wrote:yeah i got a 14900KF so the cpu part is covered,
You can say that again lol.
Harze2k wrote:ill check the pre built libmpv from the link you sent, cheers!
What site did you say had the plex fixes?
https://github.com/mitzsch/mpv-winbuild-cmake is the site. It also provides info on creating a Linux mpv build environment. Or you can use https://github.com/m-ab-s/media-autobuild_suite to create a Windows build environment but it's not MPV centric and I found builds failed more often. It's also extremely slow, a full MPV build with Plex extras takes about 30+ minutes.
There is a version with some minor fixes for Plex you can build and I can show you the web page on how to start. You basically need to set up a full Windows or Linux gcc or Clang build environment. It helps if you have a PC with a modern CPU. But if you just want the latest mpv then download the latest version of libmpv from https://sourceforge.net/projects/mpv-pl … dows/files
Yeah I saw it earlier but I didn't bother to update because after testing it doesn't seem to be any different from 4.12lite or v9. I will wait for a full v4.13 or maybe 4.14 before commenting on here.
Blackfyre wrote:Will this driver scheduled for later today promising 5x the performance increase for TensorRT-LLM for 3000 and 4000 series GPU's have an impact on us using RIFE? I assume no.
Or will this improve performance for us? Or maybe something that needs a RIFE update to utilise version 0.6.0 benefits? Not sure. Just asking.
NVIDIA states a 5x performance boost with TensorRT-LLM v0.6.0 which will be available later this month
https://wccftech.com/nvidia-brings-up-t … x-30-gpus/
Chainik wrote:how exactly "a library that leverages NVIDIA RTX architecture for accelerating large language models (LLMs)" is connected with us?
Posts found: 276 to 300 of 404