RTX 5070 ti, interpolate stuck at Running RIFE (CUDA)... perpetually

[pocketpoetry]

I've just installed Flowframe and after model download and I tried to interpolate, I'm stuck at Running RIFE (CUDA) perpetually, I suspect its got something to do with pytorch but i'm inexperienced to do any tweaking under the hood, appreciate if anyone resolved this before.

Thanks!

[n00mkrad]

Try RIFE NCNN

[Avengersman]

Try RIFE NCNN

It works with this ✅

[jacekpoplawski]

same problem with 5070

[Jinomee]

The issue is with the pytorch version and the newest 50 series card. You just gotta install the nightly build of pytorch for CUDA 12.8 and reinstall flowframes, and during the installation select System existing python + pytorch installation and it should work.

[Jinomee]

The issue is with the pytorch version and the newest 50 series card. You just gotta install the nightly build of pytorch for CUDA 12.8 and reinstall flowframes, and during the installation select System existing python + pytorch installation and it should work.

I've managed to fix this issue with Python 3.10 + Nightly Pytorch CUDA 12.8 installation, rendering 1300 frames in just under a minute.

[whitebeltmillionaire]

I followed this video to get it working 👍
https://youtu.be/1rGmDneq3i0?si=ViZYHdHgxfwtTsZu

[Chileex]

The issue is with the pytorch version and the newest 50 series card. You just gotta install the nightly build of pytorch for CUDA 12.8 and reinstall flowframes, and during the installation select System existing python + pytorch installation and it should work.

I'd just like to add that this is definitely the answer. I just bought a 5070 Ti too and was pulling my hair out getting this to work. I want to save everyone the trouble, so I'll detail what needs to be done, step by step.
Best to leave all installation paths at the default. I also suggest to read each individual step before you follow it.

• Update your GPU Drivers
• Download Python (Version 3.14.0 is working for me) from the official website and install it. I used the python install manager from there.
• Now you need PyTorch. Visit the PyTorch website and use the interactive Version selector. I used: Stable (2.9.1), Windows, Pip, CUDA 12.8, Python and it gave me this command: pip3 install torch torchvision --index-url https://download.pytorch.org/whl/cu128
  My Python installation is weird tho, maybe because I had Python installed from the microsoft store before and stuff. Anyway..
  I had to run python -m pip install torch torchvision --index-url https://download.pytorch.org/whl/cu128 --break-system-packages
  This command needs to be run in the Command Prompt aka CMD or Powershell.
• You also need some other Python Libraries. Those can be installed with pip install opencv-python sk-video imageio
  Again, I had to use a slightly different command python -m pip install opencv-python sk-video imageio --break-system-packages
• If everything went as it should, you can now install Flowframes according to the nice person I quoted. I congratulate you!

If you are as lucky as me and still get "missing module" or some other error, then there is still work to do lol. I had to also:

• Run python -m pip list -v or in your case probably pip list -v. This shows you the path pip istalled your libraries to.
• Run python - this starts a live python session in your CMD window.
  There paste import sysconfig; print(sysconfig.get_paths()["purelib"]), this apparently prints the location of where your python installation is looking for libraries.
• The two paths were different. So I made sure to copy the entire content of the "site-packages" folder from the path pip showed into the "site-packages" folder from the path that the python session showed.
• Then I finally installed Flowframes. If you installed it before and it didn't work, before completing the last steps, simply uninstall and install it again.

I can happily report that it is working now. The first run took a few seconds to start the actual generation at the "Running RIFE (CUDA)" step. Have fun :)

[Trickside1892]

* You also need some other Python Libraries. Those can be installed with `pip install opencv-python sk-video imageio`
  Again, I had to use a slightly different command `python -m pip install opencv-python sk-video imageio --break-system-packages`

Thank you so much for these steps!

In case it helps anyone else, FlowFrames reported SciPy was missing too:

pip install scipy

I also had to install the extra libraries above separately, with a no dependencies command, otherwise I was still getting errors.

pip install opencv-python sk-video imageio --no-deps

[madavvemula28-ux]

The issue is with the pytorch version and the newest 50 series card. You just gotta install the nightly build of pytorch for CUDA 12.8 and reinstall flowframes, and during the installation select System existing python + pytorch installation and it should work.

I'd just like to add that this is definitely the answer. I just bought a 5070 Ti too and was pulling my hair out getting this to work. I want to save everyone the trouble, so I'll detail what needs to be done, step by step. Best to leave all installation paths at the default. I also suggest to read each individual step before you follow it.

• Update your GPU Drivers
• Download Python (Version 3.14.0 is working for me) from the official website and install it. I used the python install manager from there.
• Now you need PyTorch. Visit the PyTorch website and use the interactive Version selector. I used: Stable (2.9.1), Windows, Pip, CUDA 12.8, Python and it gave me this command: pip3 install torch torchvision --index-url https://download.pytorch.org/whl/cu128
  My Python installation is weird tho, maybe because I had Python installed from the microsoft store before and stuff. Anyway..
  I had to run python -m pip install torch torchvision --index-url https://download.pytorch.org/whl/cu128 --break-system-packages
  This command needs to be run in the Command Prompt aka CMD or Powershell.
• You also need some other Python Libraries. Those can be installed with pip install opencv-python sk-video imageio
  Again, I had to use a slightly different command python -m pip install opencv-python sk-video imageio --break-system-packages
• If everything went as it should, you can now install Flowframes according to the nice person I quoted. I congratulate you!

If you are as lucky as me and still get "missing module" or some other error, then there is still work to do lol. I had to also:

• Run python -m pip list -v or in your case probably pip list -v. This shows you the path pip istalled your libraries to.
• Run python - this starts a live python session in your CMD window.
  There paste import sysconfig; print(sysconfig.get_paths()["purelib"]), this apparently prints the location of where your python installation is looking for libraries.
• The two paths were different. So I made sure to copy the entire content of the "site-packages" folder from the path pip showed into the "site-packages" folder from the path that the python session showed.
• Then I finally installed Flowframes. If you installed it before and it didn't work, before completing the last steps, simply uninstall and install it again.

I can happily report that it is working now. The first run took a few seconds to start the actual generation at the "Running RIFE (CUDA)" step. Have fun :)

Hey bro, I just wanted to say thank you. I bought my new dream PC with my own savings mainly for editing, and when CUDA wasn’t working I got really stressed and sad. I followed your message step by step, and now RIFE CUDA is finally working perfectly. You seriously saved me a lot of stress. Thank you so much, bro 🤍 Hope this helps others too.