Yuanbao generate movie posters, the detail capture is not quite there yet

P1 P3 are the result images, except P1 which is a front-facing shot that retained most of the details, the other posters basically don't look like me. So the image generation model has a limitation - 2D is hard to match the multi-dimensional fit. Actually, we could learn from video AI, first build an individual database, use real-time face recognition to customize the face, or allow uploading multiple photos to improve the authenticity and restoration of the character.