BUStega: A generalized coverless image steganography framework based on diffusion models and U2-Net

Image steganography has evolved along two major paradigms: traditional methods that embed secret images into cover images, and the emerging Coverless Image Steganography (CIS), which eliminates the need for carriers by directly embedding data through image features, significantly enhancing concealment. Recent CIS approaches driven by diffusion models have gained attention but face three critical challenges: (1) dependence on private prompts, which, if leaked, compromise security; (2) frequent key exchanges during transmission, increasing the risk of leakage; and (3) limited generative fidelity, leading to loss of detail in reconstructed images. To address these issues, a novel training-free CIS framework, BUStega, is proposed, combining diffusion models with U${}^2$-Net. A semantic private key is generated through optimized BLIP visual encoding and semantic distillation, improving both uniqueness and robustness. U${}^2$-Net is employed to generate adaptive embedding masks based on visual saliency, enabling more precise information placement. The integration of diffusion-based modulation and mask-guided control enhances robustness against compression and noise attacks. Comprehensive evaluations on the Stega260 benchmark demonstrate that the proposed method outperforms existing techniques in terms of security, robustness, steganographic quality, and reconstruction fidelity, highlighting its strong potential for generalized CIS applications.