Open chensav opened 2 weeks ago
To keep the code simple I completely removed batch size in this version. (also when I made this I only had GPU 2GB, so I couldn' use larger batch + mask + enhancer)
Nevertheless the converted onnx-models support batch size...
I will not implement it again, but maybe if you want to do it, look at the simple version here: https://github.com/instant-high/wav2lip-onnx Main differences I think are in DATAGEN function.
It supports wav2lip_batch_size parameter. But no face alignment, masking or enhancer
I want to run the gpu to full speed