With the help of GAN (Generative Adversarial Network), it has been possible to train the model with existing photos and generate non-existing random individuals.
- It is becoming difficult to identify which images are real and which are fake.
- Even there can be images of people who actually do not exist at all.
- The cover photo I have used is an AI-generated fake photo.
- These photographs can be used on social media platforms to misidentify, harass, etc.
In this blog post, I have mentioned a few tricks to detect a fake AI-generated image with your bare eyes.
- These tricks are not 100% guaranteed that these are fake images. But, should be enough to bring in a sense of intuition and make you alert. You need to do your own due diligence.
- With more advanced AI, I have no doubt that these tricks will be obsolete.
1. Eye Alignment
- Check if the face and eyes are center-aligned on the image.
- Even if somebody will crop the photo, check for the eye alignment always parallel to the ground.
- Even if someone will rotate the image, you can easily find it.
2. Ear or Earring
- Check if one of the earrings is missing.
- Check the ear is missing fully.
- Now the GAN models forget these things, but sure they will improve.
- Check if hairs are abruptly on air. You can see this picture to get an idea.
- Vague background, which does not represent any actual place or any place at all.
- Actual people click photos in a realistic background.
5. Irregular pupil size
- This one is a bit difficult to check.
- To know this you need to zoom in till you get to see the eye pupil clearly.
- The left one is a real child, the right girl is a fake one.
- Check if eye colors are mismatched.
- The face is deformed or another nearby person's face is deformed.
- Check for eyelid hairs, if they do not look right.
- StyleGAN2 - Official TensorFlow Implementation source code
- Analyzing and Improving the Image Quality of StyleGAN paper
- That smiling LinkedIn profile face might be a computer-generated fake
- Images credit: thispersondoesnotexist.com
- Eyes Tell All: Irregular Pupil Shapes Reveal GAN-generated Faces
Hope you learned something new today and stay alert on social media. As technology is a two-faced sword, with the advancement of technology, we need to be vigilant.
If this is the first time you are reading my article, then welcome here, I am Soumendra you can know more about me in the About me section.
Inspiration for this blog: