Post

What happened?
Install continuously failed and would not correct

Steps to reproduce

1.

Your system (OS / GPU / RAM / VRAM / etc.)
intel b660m i7 16gb RAM zotac 16gb 5060ti
Logs / full error output
Title: facefusion.py run fails with ModuleNotFoundError: No module named 'onnxruntime' (and cv2) on fresh NVIDIA/Linux install

After a clean Install + Start on Pop!_OS (RTX 5060 Ti, NVIDIA), running the Dev shell command fails immediately on import, before FaceFusion even gets to its own CLI:

eval "$(conda shell.bash hook)" ; conda activate '<app>/.env' && python facefusion.py run

Traceback (most recent call last):
  File ".../facefusion/facefusion.py", line 7, in <module>
    from facefusion import conda, core
  ...
  File ".../facefusion/facefusion/types.py", line 4, in <module>
    import cv2
ModuleNotFoundError: No module named 'cv2'

After manually installing opencv-python, the next import fails the same way:

  File ".../facefusion/facefusion/types.py", line 7, in <module>
    from onnxruntime import InferenceSession
ModuleNotFoundError: No module named 'onnxruntime'

Root cause:

Neither opencv-python nor any onnxruntime variant ended up installed in .env after the original Install step, even though the wrapper's install.js already has correct GPU-detection logic:

if ((platform === 'linux' || platform === 'win32') && gpu === 'nvidia')
{
    return 'python install.py --onnxruntime cuda';
}

and facefusion/facefusion/installer.py correctly maps cuda → ('onnxruntime-gpu', '1.24.4'), stripping any onnxruntime line from requirements.txt and installing the matched GPU package alongside everything else.

So the intended install path is correct — but on my system it seems this step either didn't run at all, or failed silently, during the original Pinokio Install. I don't have logs from the original install to confirm which, but the environment ended up with none of requirements.txt's packages installed (not just onnxruntime — cv2/opencv-python was also missing), which suggests the whole python install.py --onnxruntime cuda step never completed successfully rather than a single package being dropped.

What fixed it manually:

cd <app>/facefusion
python install.py --onnxruntime cuda

This ran cleanly and installed the full correct dependency set (onnxruntime-gpu==1.24.4, opencv-python==4.13.0.92, numpy==2.2.1, gradio==5.44.1, etc.), and python facefusion.py run worked immediately after.

Suggested things to check:

1. Confirm whether install.js's gpu === 'nvidia' branch is actually reliably triggered for newer NVIDIA GPUs (I'm on a Blackwell-architecture RTX 5060 Ti — similar detection gaps have shown up in other Pinokio launchers I've used with this card, so it's possible GPU detection silently falls through to a default/no-op path for newer cards).
2. Check whether Pinokio's Install step surfaces errors from install.py clearly if it fails partway — in my case there was no visible error at Install time, just an empty/incomplete .env discovered only when I tried to run the app.
3. Consider having install.js verify (e.g. pip show onnxruntime-gpu) after running the install command and fail loudly if the expected package isn't present, rather than silently completing.

Happy to test on request — let me know if there's a way to grab install-step logs from Pinokio for you if this happens again.
Paste the full log output here