Internet users often share photos and videos on different websites and social networks without considering the potential risks involved. That’s when so-called “deepfakes” or fake photos and videos can appear, which, although they look authentic, are actually not. To make matters worse, deepfake multimedia content is becoming increasingly difficult to detect, as the technologies used to create such content can make a photo or a video almost impossible to tell if it’s a fake, deepfake item, or if the item is authentic.
How can you recognize deepfake multimedia content? Continue reading to find out the answer to this question.
Types of Deepfakes of Multimedia Content That You Should be Aware of
The emergence of deepfakes is a fairly new phenomenon, which can be said to have surprised many. The origin of deepfake media comes from artificial intelligence technologies, such as “stable diffusion” and GANs (generative adversarial networks). Regarding the types of deepfake media, there are three main ones:
- Face replacement technologies
With the help of this technology, you can replace people’s faces. This is particularly inconvenient because unless you’re an expert and understand deepfake, and this kind of face replacement is very well done, you’ll hardly be able to recognize that it’s deepfake technology.
- Voice generators that work on the basis of AI
Don’t like the way your voice sounds? Now you can use a number of artificial intelligence-assisted voice generation technologies to produce a synthetic voice that sounds like the real thing.
- Video synthesis software
There are many apps that can generate deepfake videos by uploading a targeted photo to a video of the user’s choice. Recently, a criminal group used an unknown video generator like this to defraud a Hong Kong-based company out of dollars via a Zoom video conference. Fraudsters managed to get 25 million dollars.
How to Recognize Deepfakes and What Should You Pay Special Attention to?
A Photo Appears Inauthentic or Video Moves Unnaturally
One of the simplest ways to know if it’s a deepfake multimedia content, especially a photo, is if the photo appears to be inauthentic, which can be discerned using some indicators. In the early days of deepfake, you could easily tell that it was deepfake content, usually by a few warning signs, such as blurring around the edges, an overly polished face, double eyebrows, mistakes, or a general “unnatural” feeling about how the face fits (or rather, doesn’t fit) to the rest of the photo. However, as these technologies have advanced, it has become more and more difficult to distinguish fake photos and videos from authentic ones. But it’s still good to pay attention to blurring, distortion, and differences in facial appearance.
With videos, things are a little different. The clearest indicator of a fake video is if there are some unnatural movements in the video, which is especially visible in the movement of body parts or parts of the face during speech. Irregularities, such as when different parts of the face show different movements, can help identify deepfake videos. There are also biometric indicators, but we wouldn’t go into that, because it’s not possible to analyze biometric data with the help of free smartphone or computer applications.
Zooming in a Photo or Video
Another way you can tell whether it’s a deepfake or the real thing is by zooming in on the content, whether it’s a photo or video.
Although at first glance a deepfake photo may look quite smooth, especially since its differences are far less noticeable compared to photos that have been edited in Photoshop, you only need to zoom inside the image to notice all the irregularities. A hidden face, irregular contours, and distorted ears are just some of the visible signs of a deepfake.
In order to spot deepfakes on the video conferencing platform, experts have recommended several similar strategies. Instead of seeing the other participant in a thumbnail or gallery view, you can have a full-screen view, which will enlarge them to fill the entire screen.
The Help of Metadata
Another way you can distinguish an authentic photo or video from a deepfake is with the help of multimedia content metadata.
Of all the artificial intelligence methods that can be used to detect deepfakes, this one is the most secure and easily accessible to everyone. The easiest way to check whether it’s an authentic or a deepfake thing is to check the meta-data of the multimedia content that you suspect is a deepfake.
On a Windows computer, right-click to open “Image Properties”. Go to the “Details” tab, where you can find camera specifications, such as manufacturer, model, exposure time, ISO speed, focal length, and whether or not flash was used. A deepfake image can never have these details.
On a Mac, right-click an image and select “Get Info -> More Info” to view the image’s meta-data.
You can find several image metadata software on the Internet, which will show you even more details. Jimpl, for example, is one of the best tools out there and is completely free to use. Just upload an image that was taken with a smartphone and then view its EXIF information. Even if the location is turned off, the MCC (Mobile Content Cloud) data is always on, which means it’s connected to the SIM card service provider. Additionally, the photo’s height, width, and megapixels will be set to their maximum values, which is something that deepfake photos simply can’t replicate.
What Are the Best Network Detectors for Deepfake Multimedia Content?
On the Internet, you can find a certain number of software pieces that are used to detect deepfake content, but the number of these is limited. Additionally, most of these pieces of software will ask you to pay a fee to use them, which is something we all want to avoid in this day and age.
But there are also some free tools.
- Fake Image Detector
It’s a free tool that digs deep into the metadata and binary files of a photo to detect deepfake content. In case the photo is authentic, you will get a content message: “No error level detected”. The tool also generates a software signature, for additional authentication.
- FotoForensics
It’s another deepfake detection tool. It’s a slightly more advanced tool, which uses a very precise method of analysis in search of errors and compression levels in photos, namely “ELA (Error Level Analysis)”. If a certain part of the photo shows a certain level of error, that part is digitally modified and added to the overall photo.
Why and How Are Deepfakes Dangerous?
As we said, deepfakes are clips or images that frequently show people who have had their voices, faces, or bodies digitally manipulated to make them seem like someone else or to be “saying” something else.
Deepfakes are typically employed with malicious intent or to deliberately disseminate misleading information. They may be intended to cause someone discomfort, fear, humiliation, and discrediting. They can be used to falsely promote popular activities people love to indulge in, such as gambling. In February, scammers endorsed a phony advertisement for Valor Casino, namely one that promotes betting apps and features Indian cricket legend Virat Kohli as a profit guarantee advocate using a deepfake of the player. That’s why, if you’re an avid gambler, you should be using the services of Bla Bla Bla Studios casino sites and other operators whose reputation has never been harmed in any way and are found on specialized online casinos’ reviewing platforms.
Additionally, deepfakes have the potential to spread false information and muddle crucial topics, such as politics, through manipulating voters. For example, at the beginning of the year, voters in New Hampshire received a robocall posing as US President Joe Biden cautioning them not to cast ballots in the state’s presidential primary. The AI-generated speech sounded really authentic. The voice incorrectly claimed that casting a ballot in the primary would bar people from taking part in the general election in November, saying, “Save your vote for the November election”.
Deepfake technology can also encourage other immoral activities, like the production of revenge porn, in which women suffer disproportionately.
