Amped Software has released a significant update to its image and video authentication software, Amped Authenticate. This latest release introduces the new Reflections filter, improved deepfake detection capabilities, and additional improvements.
New Reflections Filter
The latest update enriches the Geometrical Analysis filter category by introducing a dedicated Reflections filter. This advanced tool addresses the challenges users face in identifying inconsistencies in reflections within images, particularly those generated by cutting-edge AI models such as Midjourney, Dall-E, and Stable Diffusion. Research presented at the IEEE International Workshop on Information Forensics and Security has shown that these models struggle to render reflections accurately, making this new filter a valuable asset for forensic analysis.
The Reflections filter allows users to analyze images with planar reflections, such as those found in mirrors or flat surfaces. By employing a technical approach similar to that of the Shadows filter, the Reflections filter enhances the robustness of the analysis, enabling users to establish reflection constraints effectively.
The addition of a “Feasible Wedge” feature further simplifies the process as users will be able to identify inconsistencies in reflections quickly.
Improved Diffusion Model Deepfake Filter
Recognizing the growing impact of deepfakes on information integrity, Amped Software has also improved the Diffusion Model Deepfake filter. This improvement provides users with clearer insights into the compatibility of images with known diffusion models, presenting two possible outcomes: compatibility with a known diffusion model or not. This critical information aids users in discerning the nature of AI-generated images, offering a pathway for further analysis.
It is important to note that Amped Software’s AI detection filter is designed to identify AI-generated content. The output of this filter should be regarded as a preliminary indicator to encourage further investigation rather than a definitive conclusion. A compatibility score approaching or equal to 1.00 does not guarantee that an image is AI-generated, nor does it confirm authenticity. This score represents the classifier’s output and does not imply any probabilistic significance.
To gain a deeper understanding of the complexities surrounding probability assessments, particularly concerning deepfakes, watch Amped Software’s recent webinar, which provides valuable insights on this topic.
Final Thought
By introducing new capabilities like the Reflections Filter, Amped Software equips investigators and analysts with the tools to stay ahead of rapidly advancing AI-generated image manipulation.
These improvements reinforce Amped Software’s ongoing commitment to supporting law enforcement and forensic professionals in their mission to detect digital forgeries, combat disinformation, and protect privacy. As the landscape of AI-driven media continues to evolve, Amped Authenticate tries to assist its users in staying one step ahead with advanced, court-admissible solutions.
For more details about the new features, read the dedicated blog post.
