Thomas De

12gb should be more than enough, but you can try this one if you like. https://github.com/instant-high/wav2lip-onnx I will upload latest version that also can be run on GPU soon.

@henryruhs, I'm not shure if I can help very much. You just can use the models if you like. Looking at your web page I guess we could talk german?

You should use range, not rangle

Download and copy the file to "checkpoints/vox.pth.tar" There is no need to download it everytime You have to create the folder "checkpoints" in the root of your tpsmm cloned repository

I've converted the models toONNX (dynamic axes for input size). Models are working but take only square input images Divisible by 128

> > I've converted the models toONNX (dynamic axes for input size). Models are working but take only square input images Divisible by 128 > > Hello, I just checked...

My conversion code, edit to convert model full or model half: (input has to be divisible by 128, (128x128), (256x256) ...... Takes some time, be patient... Also some warnings, but...

ONNX Model input/output shapes should be input (1,3,h,w) and output (1,3,h,w) or (batch, channels, height, width) It's a very long time ago when I coded in C++ So this is...