We propose FACE, a concept-based explanation framework that ensures faithful alignment between model predictions and concept-based representations by combining Non-negative Matrix Factorization with KL divergence regularization.