Yuanbao generated movie posters, but the detail capture capability still needs improvement.

P1 P3 are result images. Only P1 retained most details because it's a frontal face, but the other posters basically don't look like me. So the image generation model has a limitation—it's difficult for 2D to match the multi-dimensional fit. Actually, we could learn from video AI by first creating an individual database, using real-time face recognition for face customization, or uploading multiple photos to restore higher authenticity of the character.