<p>The swift evolution of digital image editing tools has led to a major debate about the authentication of content in multimedia forensics. Part of the recognition of manipulations is the detection of semantic artifacts, and also the uncovering of very subtle traces that are most of the time ignored, which may eventually lead to the understanding that alterations were made for the detection of a specific manipulation. Methods that depend on manually designed features or traditional convolutional neural networks are, in most cases, not able to effectively adapt to new or subtle variations. Our proposed framework is a multi-domain, edge-guided, and frequency as well as RGB-based model. The frequency branch is used to locate high-frequency distortions, whereas the RGB branch is used to model spatial dependencies for the detection of inconsistencies in context. The two branches make use of ResNet18 backbones with skip connections to more effectively enhance feature learning, together with an edge supervision module that is dependent on boundary feature extraction using the Sobel filter. The framework merges the features of different domains to deliver robust and distinctive representations, thus outperforming other models with an F1 metric of 96.38. The adaptability, as well as the robust localization capability, shows that there is a considerable scope of application in the field of on-the-ground multimedia forensics.</p>