With how crazy good deepfake tech is getting, I know it’s getting really tricky to tell what videos and pics are real or fake these days. As these AI-generated fakes spread across the internet, we all need to get savvy to spot the signs so we don’t get duped.
Let me walk you through the ins and outs of identifying these deepfakes. I’ll be real with you this stuff is advancing so rapidly that it’s impossible to detect every single deepfake 100% of the time. But there are a bunch of common mistakes these algorithms still make that can totally give away their computer generated nature if you know what to look for.
What is Deepfake AI?
Deepfakes use a crafty AI technique called deep learning to manipulate images, video, and audio to depict events that never actually happened. This advanced tech can realistically swap people’s faces or have them appear to say stuff they never said.
Deepfakes started as hobbyist tech on sites like Reddit, but have spread like wildfire to create viral celebrity videos, revenge porn, and political disinformation. And they are only getting more sophisticated and accessible, making it tough we learn how to indentify the signs of fakeness.
How Do Deepfakes Work?
The nuts and bolts of deepfake creation uses neural networks to analyze tons of images/videos of a person to learn their facial movements, expressions, and speech patterns. And deepfake uses that large database of images and videos to identify each and every movement, expression and talking style and these database trains AI system to create new images, videos, footages and audio that replicate the target person and mimics their actions into a new scenario.
While the outputs can look and sound shockingly realistic, deep learning hasn’t mastered perfectly replicating all the nuances of natural human appearances and behavior. Those lingering imperfections are what create cracks in the facade we can detect.
Check for Glitches Around the Face
Take a close look at any fuzziness or distortion happening around facial features. Does the mouth area look super weird when the person is talking? Since AI still can’t perfectly mimic natural facial movements and expressions, you might catch some obvious glitches around the most expressive parts of the face.
Subtle blurring or fading around the eyes, chin, and hairline is also suspicious. And watch for any pixelation those blocky digital artifacts are a dead giveaway that video was manipulated.
Examine the Facial Proportions
Here’s another easy facial feature to analyze – do the different parts of the face have natural, symmetrical proportions? Are the eyes, ears, nose, and mouth aligned in a balanced way? Or do some features look bigger, smaller, or just kinda off?
When you see asymmetrical or just wonky proportions, that’s a strong sign of a digitally generated face. Real human faces in photos and video tend to have that natural symmetry we’re all used to seeing.
Check Out How They Move
Besides analyzing the face itself, take a good look at how the person in the video moves around. Does their head seem to turn or tilt in weird, rigid ways? And do they lack those tiny natural movements we all constantly make – blinks, eye darts, little fidgets?
Also check if their shoulder and torso area moves in stiff or strange ways. If their overall motion seems oddly robotic and not smooth and natural, that’s your cue of doctored fakery.
Peep the Background
Here’s one a lot of people forget to check. But looking at the background scene can easily reveal sloppy edits. Notice if the lighting on the person looks different than the scene itself. Or if objects in the distance are weirdly warped and stretched while the foreground looks fine.
If the background seems flat, distorted, or just off, the video creators were likely only manipulating the foreground person. The untouched environment can give away the fakery.
Verify the Source
Before even inspecting the media, it’s smart to start by checking where it came from. Does it come from a sketchy, unverified social media account? Some unknown website you never heard of? That alone makes it way more likely to be fake.
And dig into the date it was posted or any timestamps shown. If those seem to clash with the apparent recency of the content itself, that’s a red flag too.
Bust Out the AI Deepfake Detectors
Don’t forget we’ve got AI working on our side too. There are now apps like Deepware that use algorithms to spot tiny facial and movement details the human eye would miss. The tech’s improving every day as researchers dig into how AI itself creates deepfakes.
While not perfect, these tools can help reveal signs of synthesis that our natural senses aren’t refined enough to catch. So bustem out to aid the search for any hints of fakery.
Conclusion
So, With how quickly deepfakes are growing and advancing, we have to stay just as savvy in learning how to stop being fooled. But if you watch for glitches, analyze proportions and motion, check the source and scene, and use tech tools at your disposal, you will become a pro at spotting the fakes and staying informed. It takes some work but we can try to stay one step ahead of the AI impersonators.
Frequently Asked Questions
How are most deepfakes created?
The majority use open-source deep learning tools like FakeApp and DeepFaceLab that use AI neural networks. Others use commercial tools like Zao or proprietary in house algorithms. Overall, deep learning gives most high quality deepfake creation.
What are the main risks and dangers of deepfakes?
Deepfakes can spread political disinformation, financial fraud using stolen identities, and harm reputations through distribution of fake revenge by creating dirty videos. There are also worry about impact on proof legal system.
Will laws and regulations be able to curb harmful deepfakes?
Potential options include requiring deepfake creators to disclose use of synthesized media and banning malicious nonconsensual deepfakes. But regulations face difficulties balancing harms, and free speech implications. Still experts believe well-crafted policies can help.
How can everyday people combat deepfakes?
Learn to identify common deepfake flaws. Use detection apps. Verify questionable media through fact checking resources. Avoid spreading sensational viral deepfakes without confirmation. And put pressure on social media platforms to quickly remove harmful identified deepfakes.