With the popularity of social networks, a vast number of posts are circulated daily. When steganographic content is posted on social networks, the focus on popular posts will draw attention away from it. Traditional generative linguistic steganography methods embed secret messages by manipulating the probability distribution at each time step using a white-box language model, which may limit generation flexibility and reduce the adaptability in the context of social networks. This paper proposes a generative linguistic steganography method based on emojis that effectively mitigates such limitations without relying on white-box models. Specifically, we first implement an emoji pre-selection strategy to identify a candidate emoji set. This set is then encoded using a mapping scheme. We select matching emojis based on the secret message and guide an LLM to generate a stego text embedding these emojis as information carriers. Our steganographic approach is well-suited for transmissions in social network platforms, enabling the generation of high-quality texts with controllable stylistic and topical features.