Temporally Coherent Human Video Deepfakes via Diffusion

Where's the code? Looks a bit suspect imo. Generating human motion without being driven by a real-world video is interesting tho. But adding audio speech as a driver. Well...