Some of the examples from the article:
Now, we need a browser extension that can do this in real-time.
That first image…why chain them all together and let the guy on the left have a gun, don’t need lines to tell it’s fake if you have even a little bit of critical thinking skills.
…guess those lines would help a large portion of the population actually…
Because having an automated way to identify what is real and what is fake can be extremely useful. That example has major red flags, but perhaps it could be that you want to check an image that doesn’t have any glaring red flags.
Not all but a lot of those blue lines seem to be quite randomly drawn.
For the hallway picture, each blue line corresponds with what should be a parallel line, along edges of square tiles or the corner of the floor and the two walls.
For the dinosaur doll, each white dot connects a feature on the dinosaur to that feature’s reflection.
For the shadows, each white dot connects a corner of the object to the corner of the shadow.
I think each of these lines is meticulously chosen.
I think the size of the lines and origin dots being used would allow the angle of any of those lines to be varied and converge or not converge. The worst case is the dinosaur foot where a different toe is selected in the mirror image than in the real. Perspective is largely numeric and you might assume AI to get that right, where AI falls down and therefore where to differentiate it is in its understanding of the objects in the image, so grossly a sixth finger or upsidedown tap. The dinosaur figurine has a phantom limb placed on its back in the mirror image, the paving in the shadows picture looks like mats laid over each other and in the soldier image they are illogically chained.
