Techniques for rendering realistic depictions of conversational gestures are provided. Embodiments include generating a data model for a first conversational gesture type, by analyzing captured video data to determine motion attribute data for a plurality of conversational gestures. Additionally, upon receiving a request to splice a gesture of the first conversational gesture type into a first animation, embodiments determine a locomotion of a first virtual character, while the first virtual character is interacting with a second virtual character within the first animation. A gesture of the first conversational gesture type is then stylized, using the generated data model and based on the determined locomotion of the first virtual character within the animation. Embodiments splice the stylized gesture into the locomotion of the first virtual character within the received animation data.