SpectraCLIP: Merging Vision and Frequency for Decoding AI-Generated Images

Given the quick development in generative AI (artificial intelligence), It has grown more challenging to determine difference between images generated by AI and authentic ones, raising concerns about misinformation, digital forgeries, and ethical misuse. This paper introduces a novel framework for identifying images generated by AI by leveraging Contrastive Language–Image Pretraining (CLIP) and Frequency analysis. CLIP provides semantic understanding by aligning visual content with textual descriptions, enabling the assessment of contextual coherence. Concurrently, frequency analysis detects anomalies in spatial frequency patterns, such as unnatural textures and spectral inconsistencies, which are indicative of synthetic generation. By combining CLIP's contextual analysis with frequency-based structural insights, the proposed method achieves enhanced detection performance. Extensive experiments demonstrate the robustness and accuracy of this approach, offering a scalable and efficient solution to the growing challenges of AI image forensics.