Deepfake technology has made great improves in the last 2 years thanks to the massive increase in video cards performance. As is already proven in many cases, combining deepfake with rendered CGI, digital double or live action double, face swap can open up new possibilities in fast realism. After several months of testing, fine tuning, mixing and combining different ai upresolution results , I have created my high resolution deepfake Ai model. The dataset contains over 7000 frames and all head directions and facial expressions. The model is based on a petrained mixed human model . So the whole training took about 3 weeks with 8 hours of work per day in nvidia rtx 3090 card. In a short scene I tried to play different facial expressions to control Ai.