How to Spot Fake AI Photos: Insights from Hany Farid
Imagine you’re a senior military officer who just received a chilling social media message: four soldiers have been captured, and if demands aren’t met in ten minutes, they’ll be executed. The only evidence? A grainy photo. What’s your first move?
Hany Farid, an expert in digital image forensics with over 30 years of experience, suggests your initial call should be to someone like him—an applied mathematician and computer scientist who specializes in authenticating digital images and videos. His work has supported journalists, courts, and governments in verifying critical evidence ranging from personal disputes to criminal cases.
The Rising Challenge
The frequency of cases involving fake or manipulated images has surged dramatically. What was once a monthly concern is now a daily issue. This spike is driven largely by two factors: generative AI’s ability to create hyper-realistic images and the dominance of social media platforms that often amplify falsehoods over truth.
We’re facing a global contest over what to trust online, and this has serious implications not just for individuals but institutions, societies, and democracies.
A Brief History of Photo Manipulation
For two centuries, photographs were generally considered trustworthy. Yet even in the 1800s, images were sometimes altered for humor or propaganda. Stalin famously had enemies erased from photos to rewrite history. With the rise of digital cameras and photo editing software at the turn of the millennium, manipulating images became easier and widespread.
Today’s generative AI allows anyone to create almost any image instantly—whether it’s four soldiers tied up in a basement or a giraffe wearing a sweater. This technology isn’t just for harmless fun; it’s being misused to create fake nudes for extortion, bogus medical advice videos, and even AI impersonators scamming companies out of millions.
How Generative AI Creates Images
Generative AI starts by taking billions of images with captions and gradually degrading them into visual noise—a random pattern of pixels. The AI then learns to reverse this process, turning noise back into recognizable images. After repeating this billions of times, the AI can generate images that match any textual prompt.
But this process is fundamentally different from how a camera captures a natural photograph, which is based on light hitting an electronic sensor.
Detection Techniques to Identify Fake Images
1. Noise Analysis
One way to detect AI-generated images is by analyzing residual noise patterns. Natural photos and AI-generated ones have different noise signatures. For example, Farid’s team examines the Fourier transform of the noise residual and looks for star-like patterns that indicate AI origin. If these patterns appear, it’s a red flag—but no single test is conclusive, so multiple checks are necessary.
2. Vanishing Points
In real photos, parallel lines like railroad tracks converge at a single vanishing point. Artists have known this for centuries. AI, however, doesn’t understand physical geometry, so its images often lack coherent vanishing points. Identifying inconsistent or missing vanishing points can reveal manipulated or AI-generated scenes.
3. Shadow Analysis
Shadows also follow physical laws. The lines extended from a shadow should intersect at the light source casting that shadow. AI-generated images often violate these rules, showing shadows that don’t converge properly. Checking shadow consistency adds another layer of verification.
Regaining Control Over What We Trust
It’s possible to separate real images from fake ones, even if it takes effort. Many of us feel like hostages to misinformation, unsure of what to believe. But we don’t have to accept that.
Here are some practical points to keep in mind:
- Tools developed by experts are becoming available to journalists, courts, and institutions to help identify fake content, indirectly supporting everyone’s ability to trust information.
- International standards for content credentials are emerging, which will authenticate media at the point of creation. These won’t solve everything but will be an important part of the solution.
- Social media should not be your primary news source. These platforms are designed to capture your attention, often by promoting sensational or false content. Approach them critically and limit reliance on them for factual information.
- Think twice before sharing information. Whether intentional or not, spreading false or misleading content makes you part of the problem. Trust and support efforts by journalists and fact-checkers who work daily to clarify truths.
The Path Forward
We face a choice: continue down a path where technology divides us, or find ways to make technology serve and support truth and connection. The decision is ours.
Rapid Fire Q&A Highlights
- What percentage of images online are fake? It varies by platform, but on sites like Twitter or X, fake content can approach 50%.
- Can AI-generated images be distinguished from those altered with filters or Photoshop? Yes, but it’s becoming more challenging.
- Are there websites for laypeople to check image authenticity? Currently, no reliable public sites exist. Beware of fake verification sites.
- Can image quality be enhanced like on crime shows? Yes, to some extent.
Learning to critically assess images and understand these detection methods empowers us to protect ourselves from misinformation. For those interested in expanding their knowledge of AI and its impact, exploring educational resources such as Complete AI Training's latest AI courses can provide valuable insights.
Your membership also unlocks:
