Flowframes Early Access Download: 1.38.2

RIFE (NCNN) works just fine but anytime I try RIFE (CUDA) I get errors [00004097] [08-03-2022 07:30:50]: [E] Traceback (most recent call last): [00004098] [08-03-2022 07:30:50]: [E] File "C:\FlowFrames\Flowframes\FlowframesData\pkgs\rife-cuda\rife.py", line 11, in [00004099] [08-03-2022 07:30:50]: [E] import skvideo.io [00004100] [08-03-2022 07:30:50]: [E] ModuleNotFoundError: No module named 'skvideo' Doesn't matter what RIFE model I use. 4.1 4.0 3.9 3.8, they all give the same error. This was not a problem with the free version 1.35 Love the program though, keep up the good work. Btw, why doesn't transparency work with RIFE (NCNN)? I only see a transparency option for RIFE (CUDA).

Upscale Anon

Image sequence inputs are broken in 1.38.2, wait for the next update pls

LTSM

I'm having issues running 1.38.2 version as well. When adding a folder with simple PNG images, it would show the following error: Canceled: Frame extraction failed. Extracted 61 frames - current.framesFolder exists: True - CurrentInputFrameCount = 0 - extractedFrames = 61. Your input file might be incompatible.

Luis Nunez

Do you have any thoughts on adding acceleration of encoding VCN for AMD? According to a recent article, it seems that B-frame has been added to VCN.

Use RIFE-NCNN-VS, it's faster anyway.

LTSM

Possibly a CPU bottleneck? Are you using NCNN/VS or the "old" NCNN implementation?

LTSM

Daniel, I did all that. I chose "Vulcan+CUDA (NCNN, Pytorch Frameworks), I installed the embedded Python runtime with all dependencies, and I still get the same error. Again, downgrading to 1.38.0 fixes it. Is there a friendly manual I need to read? Is there something I should know but I don't because I haven't Read The Friendly Manual™?

When you installed the new version did you choose "Vulkan+Cuda (NCNN, Pytorch Frameworks)" on the "Select GPU Acceleration" screen and "Install embedded Python runtime with all dependencies" option on the "Select Pytorch Installation To Use" screen? I find that fixes most problems though the extra download time is a little annoying.

Is it possible the updates don't work for me because my GPU is GTX nor RTX?

I'm using RIFE-CUDA because it's the default and I barely know what I'm doing. Which NCNN is best to use with an NVIDIA GPU? I'm upscaling AI art if it matters. Here's a folder with some frames than can render with 1.38.0 but not with either of the recent updates. https://drive.google.com/drive/folders/1fHWem0f-NO0ao9hL0eXWsiCZcX77Kfv2?usp=sharing

Appreciate the FPS limiting addition. I just have one main problem plaguing me. Interpolation will start at a high FPS whether on NCNN or CUDA, but it will slowly go down to a crawl. So for example starting at 10 FPS, to about 3-4 FPS over the course of the interpolation, which would make a 50 minute interpolation take around 2 and a half hours instead. Any clue as to why?

Jake

1) Send your input file and I can take a look if you want 2) Is there any reason you use RIFE-CUDA? NCNN is better in basically every way now

LTSM

Neither this update nor the previous one work for me. 1.38.1 works fine when I downgrade. Here's a screencap of the error. https://drive.google.com/file/d/1Q9b6RmEI5Z57Ko5ODkds3G8aP-nme7Ap/view?usp=sharing