The technologies in face recognition have undergone substantial development and have far-reaching applications in security, surveillance, and personal identification. Despite such developments, most of the existing systems suffer from inconsistent performance. This demands sophisticated models of face recognition, adaptable enough to face these challenges. We are proposing updated way with MSFF-FaceNet model that significantly enhances face recognition accuracy with advanced preprocessing and embedding techniques. Our proposed MSFF-FaceNet model makes use of feature fusion for matching fine-grained details and high-level abstractions of facial features and provides better performance in face recognition applications. Our architecture uses the FaceNet model, efficiently mapping facial images to a 128-dimensional embedding space in such a way that pictures of faces lie close together if similar and far apart if not similar. We use MTCNN for face alignment and preprocessing; this is for feature standardization of the face, which makes our model perform better. A triplet loss function ensures refinement in embeddings to several distances appropriately. Finally, we use transfer learning to fine-tune the network on our dataset using pre-trained weights. Our test results indicate that the accuracy of face recognition in real time is considerably improved in our proposed system.