We have many Intellectual Properties (IPs) of old TV shows and movies. We want to improve the visual quality of those old videos. We prepared paired data as training dataset. We trained a video to video diffusion model based on CogVideoX. Based on the output results, the model works well on those old videos. The videos become more clear and less noirse. We can see more details for some scenes.
Since the project is for commercial use, I choose a few clips and demo here.