|
Hi,
In today’s world of digital overload, we’re constantly surrounded by fake pictures and videos. For a few weeks, I was a member of the “debunking” platform metabunk, where we analyzed images and videos to determine if they were fake. Our investigations included picture analysis as well as other data such as EXIF metadata, location information, and mathematical checks to figure out what is going on.
I’ve been experimenting with different components and tools in GMIC to help with this kind of analysis, but I’d like some feedback and ideas on how to achieve more reliable results. The main goal is to determine whether a photo is original. Of course, it’s impossible to “debunk” an image showing, for example, a real person dressed as an alien, because in that case, the photo is genuine even if the subject looks fake. That distinction is important to keep in mind.
There are some online fake image detectors, but they often require a reference original for comparison. In most cases, we only have a single image and a claim to evaluate. While certain cues such as inconsistent backgrounds can reveal manipulation, I’m wondering whether any tools can reliably detect these issues.
AI now produces extraordinarily realistic pictures — for example, using tools like Forge UI or Flux AI models. You can see AI in action at "This Person Does Not Exist". There are also anti-GAN tools designed to detect GAN-generated faces.
Regarding the current trend on Social Media about “fake space pictures,” it seems many people misunderstand the issue. NASA often releases artistically rendered or composited images, which are not fake but rather enhanced. The misunderstanding is largely a communication issue on NASA’s side.
I want to use GIMP as a Analysis Tool the maybe debunk Fake Images.
Best regards, bitpen
|