A machine-learning process is used to synthesize new faces . Only three training faces were used to return out many faces. the manipulation is based on some feature face-curves. each pixel in these faces have many possibilities to perform in a combination of Red, green and blue. And according to our question for each pixel for the new features from its points of view, the nearest answer of these possibilities is selected. Then by sharing different points of view inside these new faces’ features, the synthesized face returned out without caring about the meaning of parts of any face.