The utilization of implicit representation for visual data (such as images, videos, and 3D models) had recently gained significant attention in computer vision research. A novel model steganography scheme with implicit neural representation was proposed. In this scheme, the message sender leveraged neural radiance fields (NeRF) and its viewpoint synthesis capabilities by introducing a viewpoint as a key. The NeRF model generated a secret viewpoint image, which served as a backdoor. Subsequently, a message extractor was trained using overfitting to establish a one-to-one mapping between the secret message and the secret viewpoint image. To transmit the secret message, the sender sent both the trained NeRF model and the message extractor to the receiver through an open communication channel. The receiver employed the shared key to obtain the rendered image from the NeRF model in the secret viewpoint. Consequently, the secret message was extracted by the message extractor. The inherent complexity of the viewpoint information prevented attackers from stealing the secret message accurately. Experimental results demonstrate that the message extractor trained achieves high-capacity steganography with fast performance, achieving a 100% message embedding. Furthermore, the extensive viewpoint key space of NeRF ensures the security of the steganography scheme.