This paper proposes a Motif-aware Riemannian Graph Neural Network with Generative-Contrastive Learning (MotifRGC) to address the challenges of Riemannian graph representation learning. The main issues include limited curvature diversity, numerical instability, and the lack of motif regularity capture in existing methods. MotifRGC introduces a diverse-curvature manifold through a product layer with a diversified factor, replacing the exponential/logarithmic map with a stable kernel layer. It also incorporates motif-aware Riemannian generative-contrastive learning to capture motif regularity in the manifold. The model conducts a minmax game in a self-supervised manner, generating fake motifs and contrasting different geometric views of the factors. Empirical results show that MotifRGC outperforms previous Riemannian models in link prediction and node classification tasks. The model is evaluated on four public datasets and demonstrates superior performance, particularly in capturing motif regularity and achieving better results in diverse-curvature manifolds. The approach addresses the issues of curvature diversity, numerical stability, and motif regularity, offering a new perspective for Riemannian graph representation learning.This paper proposes a Motif-aware Riemannian Graph Neural Network with Generative-Contrastive Learning (MotifRGC) to address the challenges of Riemannian graph representation learning. The main issues include limited curvature diversity, numerical instability, and the lack of motif regularity capture in existing methods. MotifRGC introduces a diverse-curvature manifold through a product layer with a diversified factor, replacing the exponential/logarithmic map with a stable kernel layer. It also incorporates motif-aware Riemannian generative-contrastive learning to capture motif regularity in the manifold. The model conducts a minmax game in a self-supervised manner, generating fake motifs and contrasting different geometric views of the factors. Empirical results show that MotifRGC outperforms previous Riemannian models in link prediction and node classification tasks. The model is evaluated on four public datasets and demonstrates superior performance, particularly in capturing motif regularity and achieving better results in diverse-curvature manifolds. The approach addresses the issues of curvature diversity, numerical stability, and motif regularity, offering a new perspective for Riemannian graph representation learning.