A method and system uses geometry-driven feature point analysis to synthesize images including for example facial expressions. Given the feature point positions (geometry) of an expression, the method automatically synthesizes the corresponding expression image, which has photorealistic and natural looking expression details.