Anyone knows what kind of (shitty) GAN models they use to upscale and interpolate frames at hstream? Cause have yet to get around on how to replicate their (laughably bad) results.