The present invention provides a method for customizing video based on viewer behaviors, by performing the following steps: receiving/preparing plurality variations of customized video related to one video template, wherein each video variation has different features including at least one of: different scenario scene, different characters, different style, different objects displaying plurality of video variations to plurality of viewers; tracking viewer behavior while watching the video and after watching the video, wherein the viewers are identified by their profile in relation to real time context parameters; grading viewer behavior based on predefined viewers target (behavior) criteria; training a neural network to select video variants having specific features per each video presentation of a specific customizable video template in relation to viewer profile and context parameters, for maximizing viewer behavior grading in relation to said video variant. applying said neural network to a given viewer profile to determine for specific video template the video features for maximizing viewer behavior grading; streaming the determined a video based on determined video features.