It’s a world full of AI-generated visuals and deepfakes and we’re just living in it.
Old deepfakes often had obvious flaws: weird faces, bad lip-sync, or odd lighting. But now they are getting better day-by-day and look/sound so sophisticated that they can trick almost anyone.
BOOM has earlier reported on how deepfakes can spread disinformation and even scam people. Here’s a comprehensive guide to help you spot deepfakes and understand how AI can also mess up at times.
But first, let’s get the definitions right of some of the most commonly used terms in the ecosystem of fakes.
Quick Facts: What Do These Terms Mean?
Morphed: Altering a part or whole of any image/video with an intention to mislead.
Cropped: Cutting out a small portion out of a larger video and sharing without full context.
AI-generated: An image, video, audio or text created using artificial intelligence usually by entering a text prompt of the required output.
Deepfake (deep learning + fake): A video or recording that has been altered using artificial intelligence and machine learning to create synthetic media that appears to be real to create the likeness of an individual. (Synthetic content is on a spectrum and they don’t always fit in clear categories)
Remember: Every AI-generated visual is not a deepfake. Every deepfake is AI-generated.
What's In A Video? Clues That Reveal It's A Deepfake
Visual Clues That Give Deepfakes Away
The ‘something's off’ feeling: Is it too good to be true or too bad to be true? This isn't just a gut feeling; it's about those tiny human details or nuances that AI misses.
Eyes and human expressions: Are the eyes blinking naturally? AI struggles with natural eye movement, blinking. Look for weird expressions that don’t really seem real.
Skin and imperfections: Look for an unnatural or uneven skin tone. Real skin often has varied textures but AI makes it look too smooth or plastic.
Background check: Observe the background carefully. At times, there can be blurred elements in the background with too much prominence given to the foreground. Also, check the shadows if they align with a light source.
Lighting: Too much, too little or uneven lighting is a major red flag.
Audio Indicators of AI Voice Cloning
So far, deepfakes are usually created with swapping faces or cloning voices. These clues can help you spot them:
Voice modulation: Sometimes, AI-generated voiceovers can sound flat with no voice modulation. Or they may also sound too robotic.
Ambient/Background noise: Even if someone is recording in a quiet room, it has some ambient noise. The lack of any background noise can be a red flag too.
Lip sync: Is the audio in sync with the video? Usually, if it’s an AI voice clone, audio and video aren’t in sync. Also, check if there are sudden jump cuts. Jump cuts usually indicate two different clips/voiceovers have been stitched together.
Click here to watch the video.
How Did We Fact-Check It? Tools to Detect AI Content
While the aforementioned clues can help you spot deepfakes, however as the technology gets better, you might need AI’s help to detect AI. Here are some free online available tools to verify whether that viral visual is legit or not.
Reverse image search: Upload the image on any of the search engines (Google, Yandex, TinEye, etc) and conduct a reverse image search to find out potential sources.
AI-Detection Tools
Use online tools such as:
Disclaimer: These tools provide helpful indicators but are not foolproof. Always corroborate findings with multiple sources.
A Step-By-Step Guide: HIVE Moderation
Step 1: Open HIVE Moderation’s website.
Step 2: Upload the image or video that you want to verify.
Step 3: The tool will analyse the visual and tell you the likeliness of it being AI-generated.
You can also add HIVE AI Detector as an extension to your Chrome browser. Here’s how you can do it:
Step 1: Go to HIVE AI Detector by clicking here.
Step 2: Click on ‘Add to chrome’ button.
Step 3: It will appear in your extensions toolbar.
Step 4: Go to the image that you want to verify and right click on it.
Step 5: HIVE AI detector will appear as one of the options. Click on it.
Step 6: It will run a scan and tell you the probability of it being AI-generated.
Also Read:AI Visuals Viral As Iran's Attack On Mossad Headquarters In Israel
Bottom Line: Question Everything
Honestly, while this checklist can be helpful; it's more important to question before you believe anything.
Question everything: Mis/disinformation is designed or packaged in a way that it plays on your emotions. No matter who has ‘forwarded’ or shared the information with you, don’t believe blindly.
Check the source: Where is the information coming from? Is it from a legit outlet, an official account, or a trusted person?
Double-check: Don't trust just one source. As fact-checkers, we often cross-check with two or more independent sources.
Spot any deepfakes you've found super convincing? Or need help with verifying any online information? Send it to BOOM’s WhatsApp tipline number: 7700906588 and we’ll verify it for you.
