In this paper, we present tiNNbre, a generative music prototype-system that reacts to timbre gestures. By timbre gesture we mean a sonic (as opposed to body) gesture that mainly conveys artistic meaning through timbre rather than other sonic properties. The system is designed and developed to be used in free improvisation and composition. Our prototype is powered by a neural network trained using a supervised learning approach on a set of sonic gestures representing the stimulus (or input) and a correspondent set of sonic gestures representing the reactions (or output). This model is then used to explore and generate new musical material. Here, we present an informal evaluation of our system, based on two use cases. In the first, the system is trained using MFCC analysis, while the second uses Constant-Q Transform spectrum. Participants were asked to rate our system’s generated audio, particularly in respect of timbre. Results showed that although tiNNbre has been tested offline, it fosters timbre-rich interaction with the sonic materials and the co-creation process. Broadly speaking, musicians preferred the results obtained using the second system other than for purely percussive gestures.