[12:50 Thu,5.March 2026 by Thomas Richter] |
As has long been predicted, the flood of AI-generated fake images and videos is continuing to increase, significantly obscuring the situation, especially concerning current events like the Iran conflict, and sowing confusion. Therefore, tools that can reliably detect such fakes are increasingly in demand—a new and quite advanced one is Image Whisperer, which is also free.
 Image Whisperer Comparison with Databases of Known AI FakesBut the capabilities of Image Whisperer go beyond that: like a human journalist checking the authenticity of an image, it first searches to see if the image has already been identified as fake or real by someone somewhere online. To do this, it searches a curated database of known fakes based on the principle of a visual fingerprint that recognizes images even if they have been cropped, compressed, or color-altered. The database is continuously updated from sources such as PolitiFact, Snopes, and Google Fact Check. Additionally, the image is searched online: when and where did it first appear, what is the very first source? This makes it particularly suitable for detecting fakes that spread rapidly via social media and can be used by journalists as well as regular users to check images before resharing them, thus avoiding the spread of AI fakes.  Image Whisperer Parallel AI Check: Where was the photo taken?After this initial check, several AI-supported detection procedures run in parallel, which attempt, among other things, to determine where a photo was taken based on landmarks, architecture, vegetation, and street scenes, in order to refute or confirm the claimed location of the photo. Furthermore, two independent systems check the image simultaneously: a commercial tool for detecting specific generators like DALL-E, Midjourney, or Stable Diffusion, as well as an academic model from the University of Naples. If both agree, the result is considered particularly reliable. In case of discrepancies, the image is flagged for human review. Additionally, it is checked whether the image contains the invisible SynthID watermark from Google&s generative AIs like Gemini or Imagen. Forensic CheckAnother check is performed forensically for telltale visual inconsistencies such as faulty shadows, perspective, reflections, textures, edges, light consistency, facial symmetry, eye details, background depth, and compression artifacts. In a further step, special attention is paid to faulty representations of people and especially faces, such as the number of fingers or distorted faces or illogical shadows. All the different analysis steps for an image are displayed transparently so that you can trace for yourself which partial evaluations from the various methods the final color-coded judgment is based on. This distinguishes 4 levels: either an image is classified as certainly AI-generated, or as unclear, probably real, or the judgment requires human verification.  Image Whisperer Judgments Although this complex procedure by Image Whisperer does not guarantee absolute protection against AI fakes, it offers an extremely comprehensive state-of-the-art toolset for detecting image fakes—especially for professionals like journalists who need to check whether an image from the internet is real or not. Given the now almost perfectly looking AI clip fakes, a version for videos, aka Video Whisperer, would naturally also be desirable. Scientific BasisHenk van Ess, the creator of Image Whisperer and a specialist in the field of open-source source verification, emphasizes  the scientific basis of the various testing methods, all of which build upon published research in the fields of digital forensics, multi-model detection, edge and light analysis, diffusion image detection, and deepfake and facial artifact analysis. Although ImageWhisperer is free, you can voluntarily support the project.  more infos at bei imagewhisperer.orgdeutsche Version dieser Seite: Fake oder nicht? Image Whisperer - neues kostenloses Profi-Tool erkennt KI-Bilder |
 |
|
|