Deepfake Detection: How to Spot an AI-Generated Face (2026)
In 2023, spotting an AI-generated face was straightforward — extra fingers, melting ears, that glassy thousand-yard stare. In 2026, it is not. The same tools that required a research lab two years ago now run in a browser tab, and the output is convincing enough that professional fact-checkers are routinely fooled on first inspection.
This guide covers what to look for, which free tools are worth using, and the behavioral signals that often catch fake accounts even when the technical tells do not. No single method is definitive — but combining visual inspection, detection tools, and behavioral analysis gives you a reliable filter.
Quick visual checklist — zoom into the photo and look for:
What Deepfakes Actually Are — and How They Are Made
The word deepfake covers two distinct types of synthetic media that are often confused. The first is AI-generated faces: images of people who do not exist, produced by generative models like GANs (Generative Adversarial Networks) or diffusion models. The second is face-swaps: real video or photos with one person's face replaced by another's. Both are now trivially easy to create. Tools that once required a GPU workstation and weeks of training data now run in a browser in seconds.
AI-generated faces (GAN / diffusion)
Models like Stable Diffusion and Midjourney generate photorealistic faces of people who have never existed. These are the profile photos most commonly used in fake social media accounts, romance scam profiles, and disinformation campaigns. The face is original — there is no real person to search for — which is why traditional reverse image search often fails to catch them.
Face-swap deepfakes
A real person's face is mapped onto another person's body or video using tools like DeepFaceLab or commercial apps. Most commonly used to create non-consensual explicit content, but also used in fraud — swapping a trusted person's face onto a video call to impersonate them.
Why detection is getting harder
Each generation of detection tools triggers improvements in generation models. Images that fooled no one in 2022 are now indistinguishable to the untrained eye. Detection tools have a half-life — what reliably catches fakes today may not catch next year's models. Visual inspection combined with behavioral analysis gives a more durable signal than any single tool.
Visual Tells in AI-Generated Photos
No generation model is perfect. Every synthetic image leaves artifacts — subtle inconsistencies that trained eyes can learn to spot. The tells change as models improve, but several have remained reliable across multiple generations of tools. The more of these you see in a single image, the stronger the case for synthesis.
Ears and earrings
Ears are structurally complex and asymmetrical. AI models consistently struggle with them — producing ears that are blurred, missing lobes, asymmetrical in unusual ways, or on faces with earrings where one earring is clearly different from the other. Zoom into the ears first.
Hair at the edges
Where hair meets a background — particularly against a light background or sky — AI-generated images often show an unnaturally smooth or blended transition. Individual strands are either too perfect or incorrectly blended into the background. Real hair at an edge has visible strand separation.
Teeth
Teeth in AI portraits are frequently blurred into a single mass, unnaturally uniform in shape and size, or feature boundary artifacts where the gumline meets the lip. Real photographs show individual teeth with slight size variation.
Eyes and catchlights
AI faces often have unnaturally symmetrical pupils and irises. Catchlights — the small reflections of light sources visible in real eyes — are frequently mismatched between left and right eye, or are geometrically impossible for the lighting shown in the rest of the image.
Background consistency
Backgrounds in AI portraits often feature repeating or warped elements — a bookshelf where books have no titles, a window reflection that does not match the room, or a background that is subtly different on either side of the subject's head. The further from the face, the less coherent the scene.
Skin texture at scale
At 100% zoom, AI skin is often either too smooth — an absence of pores and fine lines — or features repetitive texture patterns. Compare forehead texture to cheek texture: real skin varies; AI skin is sometimes suspiciously uniform.
Accessories and text
Glasses frames frequently show inconsistent reflections or pass through the face rather than sitting on top of it. Any text in the background — on signs, books, or clothing — is almost always garbled or illegible. This is one of the most reliable tells in current models.
Visual Tells in Deepfake Video
Video deepfakes are harder to sustain than still images — inconsistencies that can be hidden in a single frame become visible over time. The tells are different from photo artifacts and focus on motion, lighting, and physiological signals that are difficult to synthesise convincingly.
Blinking rate and pattern
Early deepfakes blinked rarely or not at all. Current models handle blinking, but the pattern is often wrong — too regular, too infrequent, or the blink is incomplete (the eye only partially closes). Watch for blinks over a 30-second window and compare to the natural 15–20 blinks per minute most people produce.
Face boundary flicker
Where the synthesised face meets the real neck, ears, or hair, look for a subtle flicker or shimmer — particularly in motion. This is most visible at high contrast edges, for example dark hair against a light background.
Lip sync under stress
Lip movements in deepfakes degrade under fast speech, unusual phonemes, or visible teeth. Pause the video during fast sentences and check whether lip positions match the sounds being produced. Mismatches are common at consonants like 'B', 'M', and 'P'.
Lighting direction shifts
When the subject moves their head, the lighting on a real face shifts naturally as different angles catch different light sources. Deepfake faces sometimes retain a lighting profile that does not change appropriately with head rotation — the face appears 'pasted' rather than present in the scene.
Physiological signals
Real faces show subtle colour variation in the forehead and cheeks tied to pulse — a phenomenon called rPPG (remote photoplethysmography). Some detection tools use this signal. Manually, it is not detectable — but it is the basis for several academic detection approaches.
Free Detection Tools Worth Using
No detection tool is definitive. Each has a false positive rate and lags behind the latest generation models. Use them as a second opinion after visual inspection, not as a first-line filter. A clean result from a detection tool does not confirm authenticity — it means the tool did not find the specific artifacts it was trained to find.
Hive Moderation (Photo & Video)
hivemoderation.com/deepfake-detection — One of the most widely tested free tools. Upload an image or video and receive a probability score. Performs well on GAN-generated faces and current diffusion model output. Unreliable on heavily post-processed images.
FotoForensics (Error Level Analysis)
fotoforensics.com — Shows compression artifacts across an image. AI-generated faces often show uniform ELA levels (no editing artifacts) where a real photo that has been cropped or edited would show inconsistency. Useful for catching composited images rather than end-to-end generated ones.
Illuminarty
illuminarty.ai — Trained specifically on Stable Diffusion and Midjourney output. Good at catching the current generation of AI-generated portraits. Highlights which regions of an image it considers synthetic.
Reverse face search (for stolen real photos)
If the photo is not AI-generated but is a real person's stolen image — common in romance scams — a face search engine will find where the original appeared. Upload the profile photo to FaceSift to check whether the face appears elsewhere under a different name.
EXIF metadata inspection
Real photos taken on a phone or camera contain EXIF metadata: device model, GPS coordinates, timestamp. AI-generated images contain no EXIF data, or only the fields added by the generation software. Use Jeffrey's Exif Viewer (exifdata.com) or Exiftool to inspect. Absence of EXIF on a photo that claims to be candid is a strong signal.
Not sure if the photo is AI-generated or just a stolen real one? Both are red flags — but they require different action. Run the photo through FaceSift to check whether that face appears elsewhere on the web under a different name. If it does, the photo is stolen from a real person. If it does not surface anywhere, AI generation is more likely.
Behavioral Tells — How Fake Accounts Act
Technical detection has limits. Behavioral analysis is often more reliable because it requires no tools — just attention. Fake accounts built on AI-generated or stolen photos follow patterns that real accounts rarely produce, because the person operating them is managing many targets simultaneously and working from a script.
Single or very few photos
A real person's social media account accumulates photos over time — varying angles, lighting, locations, and ages. A profile with one perfect headshot, or a handful of photos that all look professionally taken and similar in style, suggests generated or carefully selected stolen images.
No tagged photos from others
Real people appear in other people's photos — at events, in group shots, tagged by friends. An account where the subject only ever appears in their own uploads, never in anyone else's, has no genuine social graph.
Account creation date vs. activity
Scam accounts are often recently created or show a gap — created years ago but inactive until recently. Check profile creation dates on platforms that show them (LinkedIn shows join date; Twitter/X shows join month and year).
Inconsistency between photos and claimed backstory
A person who claims to be a 52-year-old doctor in Texas should have photos consistent with age, environment, and profession. AI-generated faces are often idealised — younger-looking, more symmetrical, and without environment-specific context.
Urgency to move off-platform
Deepfake video calls are computationally intensive and imperfect under scrutiny. Accounts using AI photos will often resist or find reasons to avoid live video — claiming poor connection, camera damage, or work restrictions. Pressure to move to WhatsApp before a video call is a reliable flag.
Protecting Your Own Face From Being Used
AI face generation does not require your photos to create a fake using your likeness — but face-swap deepfakes do. The more high-quality, varied photos of your face that are publicly accessible, the easier it is for someone to train a model on your appearance. These steps reduce that exposure without requiring you to disappear from the internet.
Limit publicly accessible high-resolution photos of your face
Face-swap models perform better with more training data. Keeping your highest-quality photos behind a private account removes the easiest source material. This does not prevent all misuse, but raises the effort required significantly.
Use different photos across platforms
The same photo appearing on LinkedIn, Instagram, and a dating app gives anyone a ready-made multi-angle dataset. Varying which photos you use per platform limits how much a single actor can collect.
Run a periodic face search on yourself
Check whether your face has appeared on sites you have not published it to. FaceSift scans for face matches rather than exact image copies — catching cases where your photo has been cropped, filtered, or used in a collage.
Watermark photos you publish for professional use
Visible watermarks do not stop a determined actor, but they add friction and make misuse easier to document when you find it. Semi-transparent watermarks at the corner are easily cropped; placing them across the subject's face is more effective.
Know the reporting path before you need it
If you find a deepfake using your likeness: document it first (screenshot, URL, date), then report via the platform's impersonation or synthetic media policy. Major platforms now have dedicated deepfake reporting flows. For non-consensual intimate deepfakes, contact the Cyber Civil Rights Initiative (cybercivilrights.org).
Detection checklist — run through these in order
- 01Zoom into ears, teeth, hair edges, and eyes at 100% — look for the visual tells above
- 02Check any text in the background for legibility
- 03Inspect EXIF metadata — AI images have none
- 04Upload to Hive Moderation for a probability score
- 05Run through FotoForensics for error-level analysis
- 06Run the face through FaceSift — if it appears elsewhere under a different name, the photo is stolen
- 07Check account creation date and whether the subject appears in other people's photos
- 08Request a live video call — insistence on avoiding one is a strong behavioral flag
Related guides
Check if the face is real
If the photo is a stolen real image rather than AI-generated, FaceSift will find where that face appears elsewhere on the web — often under a completely different name.
Search This Face →