The growing sophistication of fake news, particularly through manipulated images, undermines fact-checking efforts and public trust. This study critically examines the viability of Image File Properties and Image Content Analysis as features for detecting fake news. By extracting metadata and utilizing Google Cloud Vision API for content classification, the research evaluates their predictive strength using linear regression, random forest, and an ensemble model to enhance classification accuracy. Results indicate that while Image File Properties and Image Content contribute to fake news detection, their effectiveness is constrained by metadata stripping on social media platforms and contextual misinterpretations in content analysis. The ensemble approach improves detection rates, yet false positives persist, particularly when metadata is unavailable or manipulated. These findings challenge the assumption that metadata alone is a reliable indicator of authenticity and emphasize the need for forensic image analysis and AI-driven deepfake detection. Future work should address data integrity challenges, explore hybrid models incorporating network analysis, and develop cross-platform verification tools to enhance robustness. This study underscores the limitations of isolated feature-based detection and advocates for a multimodal, adaptive approach to counter increasingly sophisticated disinformation tactics.