XaiJu
n00mkrad
n00mkrad

patreon


Flowframes Early Access Download: 1.38.1

This update primarily focuses on some things that were broken with the new VapourSynth implementation of RIFE (RIFE-NCNN-VS).

It now respects the GPU ID set in the settings (NCNN). However it currently can't do multi-GPU, so if you specified multiple IDs, it will only use the first one.

Trimming now also works with RIFE-NCNN-VS.

In addition, the VRAM/RAM usage should now be lower at very high (4K+) resolutions.

Full changelog on Discord and in Flowframes.

Comments

DM me the full error and I'll look into it

LTSM

Well, in 1.38.2 I got a different error after OS restart when I use VapourSynth - it throws KeyNotFoundException when I click on interpolate, and was fine before the OS restart

rocketraccooon

Pytorch no longer gets installed by default because RIFE NCNN is much much smaller and (nowadays) faster. However the message shouldn't pop up every time, this has been fixed in 1.38.2. If you still want Pytorch you can enable it in the installer and it should work.

LTSM

I believe the early access installers are not as complete as releases - they seem to skip necessary framework installation. The result is, it works right after installation, and is guaranteed to fail after an OS restart.

rocketraccooon

Every single time I update and/or launch the app I get the Python error thrown. I can't see where/how to fix this. Any suggestions?

Glen McBride

I had to downgrade to 1.38.0. The update couldn't recognize PNG files exported from a colab I'm using called IllusTrip3D that were then upscaled with RealESRGAN.

Unfortunately I could not fix the issue with a re-install. In fact my antivirus even flagged it as ransomware and deleted the exe which was annoying, since it's never happened before.

Jake

Will do, thanks!

Jake

Use RIFE NCNN, it should be faster and FPS limiting works (just not in the VS implemention)

LTSM

I use RIFE-CUDA because FPS limiting doesn't work with NCNN-VS atm. I always use H265 with NVENC. Perhaps when installing something went awry, I'll try to reinstall it later and see if the issue persists. The most evident part is that when It starts to interpolate, it starts at below 1 FPS and rises up to about 4-5, where it's usually 9-10 from the start.

Jake

What AI are you using? RIFE-NCNN-VS? I have not changed anything about the executable itself since 1.38.0. Maybe you switched to a slower codec and now you're CPU-bottlenecked?

LTSM

It's definitely on the right GPU, after re-installing 1.38.0 the issue is fixed. It was simply not utilizing my GPU or CPU much at all, 12900k/3090. Might just be a bug, I used the same settings.

Jake

Do you have an integrated GPU? Maybe it's now running on the wrong GPU since it previously would use the default one?

LTSM

Question, after the update my interpolation speeds have drastically lowered with my GPU and CPU both spiking like this, not sure why. https://gyazo.com/ae92da73ae2b255e77325dc65c90d73f

Jake

Shouldn't be too hard, I just haven't look into it much yet. Probably next month.

LTSM

Thanks for the update. Do you know when FPS limiting will be fixed with RIFE-NCNN-VS?

Jake


More Creators