In software like rope and facefusion there is a function to increase the resolution of the underlying face replacement model (inswapper 128 -> 256, 384, 521...)
The function cleverly uses pixel interpolation and increases resolutions, increasing generation speed (128 -> 256, x4 generation time).
Will it be possible to use this technology for Live Portrait model as well? Increase resolution from 512 to 1024
I'm not a programmer myself and I don't know all the intricacies of AI models, I don't see the difference between .pth and .onnx, maybe in this case it's not possible, but I would be very happy with this integration :)
In software like rope and facefusion there is a function to increase the resolution of the underlying face replacement model (inswapper 128 -> 256, 384, 521...)
The function cleverly uses pixel interpolation and increases resolutions, increasing generation speed (128 -> 256, x4 generation time).
Will it be possible to use this technology for Live Portrait model as well? Increase resolution from 512 to 1024
I'm not a programmer myself and I don't know all the intricacies of AI models, I don't see the difference between .pth and .onnx, maybe in this case it's not possible, but I would be very happy with this integration :)
Examples of pixel boost work:
