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 that retained most of the details, the other posters basically don't look like me. So the image generation model has a limitation - it's hard for 2D to match the compatibility of multi-dimensional features. Actually, it could learn from video AI by first creating an individual database, using real-time face recognition to customize faces, or uploading multiple photos to restore higher authenticity of the character.
原文表示