Yuanbao generates movie posters, but the detail capture isn't quite there yet

P1 P3 are the result images. Except for P1, which is a frontal shot and retained most of the details, the other posters basically don't look like me. So there's a limitation with the image generation model - it's hard for 2D to match the multi-dimensional fit. Actually, we could learn from video AI and first build an individual database, use real-time facial recognition to customize the face, or allow multiple photo uploads to restore higher authenticity of the character.