“Perfecting” perfection - uncovering problematic celebrity photoshopping

Jennifer Anniston, Angelina Jolie, and Beyonce have long been regarded as some of the most beautiful women of our time.

And yet the truth is those regarded as “perfect,” still don’t meet the bar of perfection for those that seek to use their images.

What does this say, when even “perfection” isn’t quite perfect enough?

What follows is an exploration using a new AI tool that could provide a much needed weapon in the fight against the increasingly unrealistic portrayals of beauty in media and marketing.

The tool is called a FAL Detector, and was developed by UC Berkeley and Adobe Research. The AI model the tool leverages is a binary classifier using a Dilated Residual Network, trained on images that had been warped using Adobe’s “Face-Aware Liquify” feature. The model outputs a  heatmap of the locations on the face where this tool has been used. The closer to red, the more manipulation of the image at that location. 

The AI also attempts to “fix” the image, or reverse the use of the Face-Aware Liquify tool. Some of the following examples include these AI “fixed” versions (only cases where the fixed version looked noticeably different).

“Perfecting” Jennifer Aniston

Jennifer Aniston has long been admired for her beauty. To many women, her features are iconic. Sadly, the decision makers at magazines, ad agencies, and even the press regularly photoshop her look, making consistent editorial decisions about what parts of Jennifer’s face don’t match their idea of perfect beauty.

We examined 20 Magazine Covers featuring Jennifer Aniston with a close shot of her face. Of these, the AI picked up clear use of the “Face-Aware Liquify” photoshop tool in 50% of the samples. The most frequently edited parts of her face in the 10 examples we found were:

1. Jaw (6 out of 10 images)

2. Chin (6 out of 10 images)

3. Lower Lip (3 out of 10 Images)

Examples:

“Perfecting” Beyoncé

Even Queen Bey can’t escape the critical eye of magazine photo editors. Admittedly, it was actually much more difficult to find clear examples of use of the “Face-Aware Liquify” photoshop tool on Beyoncé. This is for a few likely reasons:

1. For some reason, there seem to be fewer magazine covers of Beyoncé with a focus on her face. Many are full body or in costume. The model works best on closeups of faces.
2. The model was trained on faces from the Flickr image dataset. These photos skew “Western” as noted by the paper authors, which likely means there are fewer black faces in the training data, potentially leading to less accurate results on Beyoncé magazine covers looked at in this experiment. 
3. Beyoncé is just so iconic, almost no one dares edit her face? True Queen Confirmed?

In the few examples where AI found there to be editing, Beyoncé’s eyes, Nose, Chin, Lower Lip, and the corners of the mouth were the primary locations.

“Perfecting” Angelina Jolie

Similarly to Jennifer, Angelina is also often the victim of aggressive photoshopping. In the 20 samples of Magazine Covers we looked at of Angelina Jolie, roughly 50% had clear evidence of using the “Face-Aware Liquify” photoshop tool.

The most frequently edited parts of her face in the 10 examples we found:

1.  Jaw (5 out of 10)

2. Corners of the Mouth (3 out of 10) 

3. Lower Lip (2 out of 10)

Examples:

Looking Forward

New and better tools for photo and video manipulation are here, and only improving by the month. These tools will undoubtedly be leveraged to their maximum potential to create perceptions in you, the viewer.  

Without tools to help us understand how the images we see are manipulated, we have little hope in understanding the motives and goals of those who create manipulated content presented as unmanipulated.  Media and content literacy can be improved when the curtain is pulled back, the manipulations made apparent. Even the “perfect” aren’t perfect enough.