An artificial intelligence app that visualizes what you looked like in ancient Egypt as a punk rocker or a fugitive from the Wild West has exploded.
People across social media share face and body versions edited by the MyHeritage AI time machine to look like archetypal figures from different eras in history.
So what do politicians look like: Vikings, ancient Greek warriors, cowboys, 1960s hippies, pirates?
Newsweek We processed and examined images of former President Donald Trump, President Joe Biden, Rep. Marjorie Taylor Greene, Florida Governor Ron DeSantis, and Rep. Alexandria Ocasio-Cortez.
donald trump
1 of 5

MyHeritage AI Time Machine™
Joe Biden
1 of 5

MyHeritage AI Time Machine™
marjorie taylor green
1 of 5

MyHeritage AI Time Machine™
Ron DeSantis
1 of 5

MyHeritage AI Time Machine™
Alexandria Ocasio Cortez
1 of 5

MyHeritage AI Time Machine™
MyHeritage, which developed this feature, is an online genealogy company that started by allowing users to create their family tree online. In recent years, the company has expanded to use his AI to bring historical photos to life and animate them. Now users can use his AI on their faces to see what they looked like in the past.
This AI works using text-to-image technology. This is unlike the basic filter art you experience on Snapchat, where the app puts dog ears on your head and distorts facial features.
“MyHeritage AI Time Machine utilizes text-to-image technology based on Stable Diffusion and is licensed from Astria,” said Sarah Vanunu, director of public relations at MyHeritage. Newsweek.
“We use many photos of one subject to build a model that can portray the same person in different poses and lighting conditions than the original photo,” she said. She “then uses a series of predefined themes to composite the model with various thematic motifs to create photorealistic images.”
Text-to-image AI uses deep learning neural network algorithms to interpret text descriptions (“Vikings” or “cowboys” in this case) and edit the input images to create realistic images. increase. Algorithms are often trained by scraping vast amounts of data from the internet to learn what a “Viking” should look like.
“The quality of the output you receive is highly dependent on the quality of the photos you upload. The more photos you upload and the more variation you have in your photos, the better the results,” says Vanunu.
As you can see in the image above, the results aren’t always perfect. Marjorie Taylor Green’s left arm appears bent backwards in her ancient Greek version, while Joe Biden’s cowboy version has an inhuman face.
Several AIs of this ilk are available online, including DALL E 2, a popular method for generating AI art from language descriptors. The results of these AI-powered experiments were shared on social media, resulting in a variety of beautiful images that many could not distinguish from man-made ones.
One strange appearance from DALL-E is an image posted by Twitter user @supercomposite. called a “robe”. Loab is an image of a red-cheeked, melancholy woman that kept popping up in his AI art in a variety of terrifying ways.
🧵: Found this woman I call Loab in April. AI recreated her easier than most celebrities. Her presence is persistent and she follows every image she touches. CW: Please take a seat. This is a true horror story that takes a sharp turn to spooky. pic.twitter.com/gmUlf6mZtk
— Super Composite (@supercomposite) September 6, 2022
@supercomposite wrote in a Twitter thread, “She’s an island that appeared in the latent space that I don’t know how to identify with a text query.” “But she said Loab was as powerful a convergence point for AI as it was for verbal concepts.”
However, MyHeritage AI Time Machine™ only uses the user’s face, so there have been no cases yet of anything spooky popping out of the image, aside from the odd similarity to photos of older family members.
“Many people have commented on the family resemblance they see between themselves and their AI images,” said Vanunu.
Any tips on which science articles Newsweek should cover? Got questions about AI? Please let us know at science@newsweek.com.