search

Nick088Official / FaceFusion-Online-Ports

FaceFusion Online Ports on Google Colab, Hugging Face Space & Kaggle
Other
50 stars 18 forks source link

Video merge at end never works in lightning ai port #30

Closed xd003 closed 1 week ago

xd003 commented 4 weeks ago

I have noticed a very critical problem in the lightning ai port because of which i am unable to use. Everything i start the process in facefusion, the end step of merging the video always fails.

image

The very same source/target with exact same settings work totally fine on google colab, so most probably something specific causing this issue on lightning ai port

xd003 commented 4 weeks ago

Okey i found the issue. Only libx264 output video encoder works in lightning ai while in colab, other output video encoders worked as well. The video merge process is totally done with ffmpeg from what i have read

If lightning.ai is only supposed to support libx264 then feel free to close this issue, I don't know the technicalities behind it. Just thought to mention it here

Nick088Official commented 4 weeks ago

Could you share both of your inputs so I could test too?

xd003 commented 4 weeks ago

Could you share both of your inputs so I could test too?

It should be reproducible with any input. I searched bit more on this, there are are amd and nvidia specfic video encoders so possibly they only work on those specific hardware

image

Also is it normal that everytime i execute the Run UI cell, it seems to be installing packages again each time. Since the storage is persistent, should't it launch the UI straight away ?

Nick088Official commented 4 weeks ago

Could you share both of your inputs so I could test too?

It should be reproducible with any input. I searched bit more on this, there are are amd and nvidia specfic video encoders so possibly they only work on those specific hardware

Both Google Colab & Lightning.ai use NVIDIA GPUs, so it would be kinda weird it works for Colab but not Lightning.ai, It would be way easier for me if you just give me the inputs you tested with, as I haven't had reports of this before.

image

Also is it normal that everytime i execute the Run UI cell, it seems to be installing packages again each time. Since the storage is persistent, should't it launch the UI straight away ?

Check https://github.com/Nick088Official/FaceFusion-Online-Ports/issues/25 , lightning.ai uses by default cuda 12.1, but FF needs cuda 12.4, and for some weird reason when I update the version, it doesn't seem to persist, so i made so that everytime it updates to cuda 12.4 before running the UI, as I even asked in their Discord Server and wasn't able to find any faster solution, however it takes only 1-2 mins.

Nick088Official commented 1 week ago

Going to close this as there's no news about it.