<![CDATA[The rapid advancement of artificial intelligence (AI) in the field of image generation has raised new challenges for digital content authentication and validity. AI-generated images are often indistinguishable from real photographs, creating potential risks of misuse in disinformation, visual manipulation, and copyright infringement. This study proposes the integration of an invisible watermarking method based on a hybrid of Discrete Wavelet Transform (DWT) and Singular Value Decomposition (SVD) directly into the AI-image generation pipeline. The system is developed end-to-end with three main components: a Translator API to support Indonesian text inputs, an AI-image generator to create images from descriptive text, and a watermarking module to embed and extract hidden watermarks automatically. Experimental results confirm that the visual quality of watermarked images was preserved, with PSNR values consistently above 35 dB and SSIM ≥ 0.95, indicating that the watermark is imperceptible to human vision. Watermark extraction evaluation achieved a position accuracy of 59.43% after normalization and a subsequence accuracy of 80.20%, demonstrating reliable recognition of the embedded watermark sequence. Robustness tests under common manipulations such as JPEG compression, rotation, cropping, and noise addition showed that the watermark remained detectable, although accuracy decreased under extreme cropping. These findings demonstrate that the hybrid DWT–SVD method is effective for ensuring the authenticity of AI-generated content without compromising visual quality, while offering novelty through its integration into the generative pipeline and its support for local language inputs.]]>
Copyrights © 2026
