<p>Neural Radiance Fields (NeRF) have been widely applied due to their powerful implicit representation capability for 3D scenes, yet their high training costs have made the copyright protection of NeRF models increasingly important. Existing watermarking methods mostly rely on complex joint optimization, which makes them difficult to adapt to pre-trained models; additionally, their single watermark modality limits the flexibility of practical applications. To address these issues, this paper proposes a NeRF watermarking algorithm based on the Least Significant Neurons (LSN) and constructs a standardized framework supporting multimodal watermarking. Specifically, the method mines the functional domain redundancy inside the NeRF network, identifies and prunes low-contribution neurons according to the weight L2 norm, and records their positions as the secret key. Under the parameter masking mechanism, the backbone network parameters are fixed, and the pruned redundant neurons are reconstructed as an integrated unit to embed and express implicit watermark information. The proposed scheme requires no access to the original training data during the watermark embedding stage and also does not require the participation of the creator. It can directly perform posterior watermark embedding on already trained NeRF models. Meanwhile, the lightweight implicit watermark structure possesses rich expressive capacity and can flexibly carry multimodal copyright information according to practical requirements. Experimental results demonstrate that the proposed method achieves accurate and robust watermark extraction while maintaining high-fidelity rendering quality.</p>