Photoshop Detector AI Is Useless

Photoshop detector AI called FALDetector aims to detect facial edits that warps the faces to make chin or jaw looks thinner, or the forehead smaller. However, it is not really accurate and consistent as it look at the individual pixels too much, as different resolution of the same person can result in different predictions with this AI model.

FALDetector

Today's Sponsor is Infinite Red
Infinite Red consulting handles your mobile, web, and AI needs

This video is supported by the kind Patrons:
🙏Marc Schwyn, Mazen Alotaibi, Sascha Henrichs, Jake Disco, Peter Davidowicz

Support me on Patreon if you hope to see more:

or by becoming a member instead (same perks!):

[Music] Steaminwaffles - Blue

[MaximilianonMars]

50mm is usually stated to be "the eye's focal length" meaning images appear undistorted. This is affected by the camera's sensor size, if it's less than full frame (1:1) then the factor it is smaller by multiplies the lens' focal length, example: 50mm lens on full frame = 50mm, on ¾ frame = 67mm.

[jcims]

Awesome job dude! Obviously you put a ton of work in this one, and I really like hearing both points and counterpoints of papers.

[TheOneThreeFour]

There's more so an ideal focal length that looks closer to reality than there is a rule that a wider focal length makes people look better.

Sub 50mm, the face begins to look so stretched and thin that it becomes weird, especially once you get to the point of fish-eye warping. Portraits on 18mm don't tend to compliment the face very well.
Above around 100mm, the face begins to flatten out like a pancake and becomes rather wide.

For portraits, people tend to say 50-85 is where you want to be to make the human face look most pleasing, but also all of this kinda goes out the window once angles and distance from the camera are introduced, as in a shot on a 400m lens from a few hundred meters away, a person can look rather fine. The foreground/background compression effect only really comes into play if you're comparing a shot on two different focal lengths where the camera *moves* to keep the subject the same size in the frame.

[joshuascholar]

That makes sense. If the focal length is narrow, then there will be blurred pixels that are out of focus. But the algorithm thinks that pixels on the face are blurred because they were warped.

[LightTheMars]

Project links in the description are invalid, copy/paste mistake?

[antonharboe_]

good video man! Sounds really funny :)

[GantMan]

This explanation and science is what I need in my life!

[subwayz_qt]

Someone should try this on "Gigachad"

[darrellm]

What if an image was liquefied, then taken a photo of from another camera, then printed on paper and then taken from another camera again? 🤣🤣

Do you think by then it would be almost impossible to detect?

[BrainPermaDeD]

Good that I am not into that influencer upload BS.