#Deepfakes are everywhere, but #DigitalForensics investigators are fighting back:
-
@AudeCaussarieu
J'aurais jamais pensé que les points de fuites pouvaient être une façon de vérifier la véracité d'une image. 🤯@mathieu_ @AudeCaussarieu fondamentalement, les AI generatives fonctionnent a l'echelle des pixels d'une image (ou video), et n'ont pas vraiment un model abstrait d'une scene (pas de plan 3D en tete)
C'est pour ca aussi sur les model plus vieux les doigts surnumeraires, les 3e mains mal placees, etc. Meme-si ces erreures la sont plus facilement corrigibles.
-
#Deepfakes are everywhere, but #DigitalForensics investigators are fighting back:
@FabMusacchio in the first picture, the lines may be off by measurements inaccuracies. a mirror image depends on a perfect mirror on the physical mirror. and the shadows may be off by camera lens distortion.
-
#Deepfakes are everywhere, but #DigitalForensics investigators are fighting back:
@FabMusacchio@mastodon.social If you think floor tiles are a good indication for parallel lines, you need to watch more CyFy videos
All jokes aside, this is a very good trick to spot AI content, and it still works with the most recent models. -
@NatureMC @ilusenn @FabMusacchio Definitely not a fan either but I am aware of where this one leads (or, at least, should).
@JonnyT I like using redirect checkers like WhereGoes, RedirectDetective, etc which I've found to be quite convenient
-
@ghoppe @leadore @leah @f4grx @FabMusacchio While that is technically true, the splay of shadows on Earth is negligible and nearly imperceptible in real life. In the example image it's as if the sun is a lamp a few metres away.
@seachaint @leadore @leah @f4grx @FabMusacchio in most situations, shadow perspective is apparent when the camera is pointing towards the sun. It’s different if camera is overhead.
-
@mathieu_ @AudeCaussarieu
Je suis quand même un chouilla réservé sur la première photo. Ok, elle semble générée par IA, mais les lignes de fuite sur lesquelles l'analyse se base sont un peu courtes. Difficile d'extrapoler sur un petit segment qui n'est pas nécessairement représentatif ni exact.
La méthode est intéressante mais pas toujours applicable@YannC @mathieu_ @AudeCaussarieu je crois que la chaîne qui rentre dans le bras gauche du soldat sur la gauche est un meilleur indice de fake que la perspective inexacte
-
@JonnyT I like using redirect checkers like WhereGoes, RedirectDetective, etc which I've found to be quite convenient
@h3mmy Why should I do that work: I only click and boost posts with real links, that's it. So easy.
@JonnyT @ilusenn @FabMusacchio -
R relay@relay.infosec.exchange shared this topic
-
#Deepfakes are everywhere, but #DigitalForensics investigators are fighting back:
@FabMusacchio damn, this is why they taught us math? To find the deep fakes posing as truth?
-
@mathieu_ @AudeCaussarieu
Je suis quand même un chouilla réservé sur la première photo. Ok, elle semble générée par IA, mais les lignes de fuite sur lesquelles l'analyse se base sont un peu courtes. Difficile d'extrapoler sur un petit segment qui n'est pas nécessairement représentatif ni exact.
La méthode est intéressante mais pas toujours applicable@YannC quand même il y a largement assez de pixel !
-
#Deepfakes are everywhere, but #DigitalForensics investigators are fighting back:
@FabMusacchio Perhaps they came to arrest the tiler?
-
#Deepfakes are everywhere, but #DigitalForensics investigators are fighting back:
@FabMusacchio Someone should do this with moon landing videos
-
#Deepfakes are everywhere, but #DigitalForensics investigators are fighting back:
@FabMusacchio I would be more tolerant in the first example with the soldiers, because many builldings are not build perfectly straight or remain like that. If there is only a small error, it would not be a strong indicator for a fake. Of course, this doesn't apply to the other examples.
-
@FabMusacchio to be fair, the first one the perspective doesn't 100% mean it's fake. It could be that some builder somewhere did a shite job of the tiling.
@nihilarc
Not just that: lens distortion, uneven lighting, confusing shades, etc. - we'll see lots of people claiming something is genai since it's not perfect, but... the world is not perfect. People are already getting called out for AI writing when it's just their style.
@FabMusacchio -
#Deepfakes are everywhere, but #DigitalForensics investigators are fighting back:
@FabMusacchio Also, what's with the chains? And look at the left thumb on front and center guy.
-
As a photographer who is very fond of extremely wide angle lenses, I can tell you these tests will often fail on real images, as well. But the points made are valid...
@Walrus @nihilarc @FabMusacchio
Well, and besides, in a photograph with shadows, what's important is not that the lines don't converge, but that the shadows are basically at different angles, which means (would mean, in the case of a real photo or a render) that the light source is located very close.
Shadows from an “infinitely” distant sun should be parallel (well, on the scale of a single city square).
-
#Deepfakes are everywhere, but #DigitalForensics investigators are fighting back:
oh this is a neat trick, I usually look for distorted writing and other small detail, but latest AI models are getting better at those
-
#Deepfakes are everywhere, but #DigitalForensics investigators are fighting back:
@FabMusacchio you forgot to poison your data , you post they learn
-
#Deepfakes are everywhere, but #DigitalForensics investigators are fighting back:
@FabMusacchio I don't know, I've seen some pretty bad tile jobs
