As deep learning technologies continue to advance, Artificial Intelligence Generated Content (AIGC) has increasingly become a prominent area of research in both academia and industry. AIGC technology facilitates the creation of content that is highly relevant to the input information, with applications across various domains. However, the widespread use of AIGC also brings about risks related to information security. This paper introduces a watermarking injection and detection framework based on proprietary image generation models. The framework involves dividing the codebook of a trained VQGAN model into red and green lists, and then applying watermark injection and detection based on these lists to the encoded and quantized images’ features. This watermarking method does not require additional training and can be seamlessly integrated into existing image generation models. Furthermore, we also propose a novel watermark detection method that allows for the direct calculation of whether watermark information is present in an image through statistical tests of encoded features. We conducted watermark injection and detection metric tests on the public datasets CelebA-HQ and ImageNet-1k. The experimental results validate the effectiveness and robustness of the proposed watermark injection and detection technology.