Here are 4 ways to tell if an image is AI-generated, as deepfakes of celebrities and politicians spread across the internet
- Fake images of Donald Trump's arrest and Pope Francis in a coat recently fooled the internet.
- Images generated by AI programs like Midjourney, DALL.E2, and Stable Diffusion are on the rise.
In recent months a number of deepfake images of major figures in unlikely scenarios, such as a viral picture appearing to show the Pope in a stylish white puffer coat and a bejeweled crucifix, have circulated online.
Like many such images, the picture of the Pope was actually made with an AI program called Midjourney, which David Holz founded last year.
The program, which creates images based on textual descriptions provided by users, has been used to produce misleading images of well-known figures including some of former president Donald Trump being arrested.
In late March, Midjourney suspended free trials "due to a combination of extraordinary demand and trial abuse," Holz said at the time.
Yet that does not mean an end to fake images, according to Henry Ajder, an AI expert and presenter who is on the European advisory council for Meta's Reality Labs. He said tools such as OpenAI's Dall. E 2 and Stable Diffusion had this capability.
"The only way that realistic fakery has been possible in the past to the level we're seeing now daily was in Hollywood studios," Ajder said. "This was kind of the cream of the crop of VFX and CGI work, whereas now many people have the power of a Hollywood studio in the palm of their hands."
He warned that the consequences of deepfake images will range from fake news about politicians to nonconsensual pornographic images.
For instance, in April a face swap app called Facemega was used to promote a sexually suggestive ad using actor Emma Watson's face.
However, it's not just the "bombastic fakes" people need to worry about, Adjer said. The more subtle ones like Pope Francis can "slowly just chip away at our trust in visual media and make it harder to navigate the truth."
He and another expert offered four tips to help distinguish AI-generated images from the real thing.
Some AI-generated images have a "plasticky" appearance
One telltale sign that an image was on Midjourney is a "plasticky" appearance, but the platform may iron out this issue as it develops.
Ajder said Midjourney was a tool developed with artists in mind: "A lot of the images have a very stylized, almost smooth kind of shiny, plasticky appearance."
Although this isn't consistent with other AI platforms, it's something to keep an eye out for.
Look out for aesthetic inconsistencies
Ajder pointed out that AI programs generally struggle with "semantic consistencies," such as lighting, shapes, and subtlety.
Some examples include checking whether the lighting on a person in an image is in the right place; whether someone's head is slightly too big; or even over-exaggerated eyebrows and bone structure.
Other inconsistencies include smiling with lower sets of teeth in an image because usually "people smile with their top teeth, not their bottom."
Not every single image will have these signs, but they're useful pointers.
Alexey Khitrov, founder of biometric security company ID R&D, said the image of Pope Francis is the "artifact of something that's completely unnatural," and contained some "physically impossible" features.
The crucifix the Pope appeared to be wearing in the image had a chain is attached to only one side, for example.
Other errors included the strange shape of his ears as well as the distance between his glasses and their shadow on his face.
Context is key
Aesthetic factors are not always enough to identify deepfakes, especially as AI tools start to become more sophisticated.
Khitrov advised questioning suspicious images: "Try to do a search on the image like you're doing a search on the information that you're receiving."
Ajder agreed context was critical, making it worth trying to find an "authoritative source."
"We need to be aware that if something seems outrageous or sensational, there's a good chance that there might be something awry. In that context, it's about going to the organizations that have known capacity, for fact-checking and verification."
He advises asking questions like: "Who's shared it? Where has it been shared? Can you cross-reference it to a more established source with known fact-checking capabilities?"
Try a reverse image search
If all else fails, Ajder suggested using a reverse image search tool to find the context of an image.
"If I did a reverse image search on the Trump getting arrested images, it might take me to all of the news websites where it's been shared in articles. So it's essentially a way to sort of trace back [the image] or like a mind map coming off that image."
- 10 cafes in Bangalore offering the best of ambiance and cuisine
- Centre's fiscal deficit at Aug-end touches 36% of full-year target: CGA
- Domestic occupiers keep office space demand high in 2023
- Nifty, Sensex bounce back on Friday but experts see resistance at higher levels
- Temperature's Toll: Hospitals see a surge in drug and alcohol abuse-related visits on hotter nights, study reveals