Deepfake photos are no longer a niche problem. Face swaps, AI portraits and edited scandals show up daily in social feeds, dating apps and news cycles. This guide is a practical, repeatable workflow to detect deepfake photos using free tools you can run in your browser.
What counts as a deepfake photo
A deepfake photo is any still image where a face, body or identity has been swapped, generated or significantly altered by AI. That includes face swaps on real photos, fully synthetic portraits, and "undressed" or "aged" versions of real people. The common thread: the image is presented as real, but the person or scene never existed that way.
The 6-step deepfake detection workflow
- Inspect the face zone. Zoom in on the eyes, ears, teeth and hairline. Look for asymmetric earrings, fused teeth, melted glasses, and hair that smears into the background.
- Check lighting and shadows. A real photo has one consistent light source. If the nose shadow goes left but the neck shadow goes right, you are likely looking at a composite.
- Read the metadata. Drop the file into the EXIF Viewer. Real camera photos carry rich EXIF; most deepfakes do not.
- Run an AI detector. Use the AI Image Detector for a probability score and a region-level breakdown.
- Reverse search the image. Open the Reverse Image Search to find the original photo the deepfake was built on.
- Cross-check with a tamper heatmap. Use the Tamper Heatmap to highlight pasted faces or edited regions via compression analysis.
Visual tells that still work in 2026
- Hands and fingers. Extra fingers and warped knuckles still appear in many generated images.
- Teeth and gums. Look for fused teeth, missing gum lines, or unnatural symmetry.
- Ears and jewellery. Mismatched earrings, melted cartilage and floating piercings are very common.
- Text in the background. Signs, license plates and tattoos often turn to gibberish.
- Skin texture. Overly smooth skin paired with razor-sharp eyelashes is a classic AI signature.
How to handle screenshots and social downloads
Instagram, TikTok and X strip almost all metadata on upload. That makes EXIF less useful for content found in feeds. Compensate by relying more heavily on reverse image search and the tamper heatmap, which both work on re-compressed downloads.
When to escalate
If the deepfake targets a real person — non-consensual intimate imagery, political impersonation, or identity fraud — escalate to the platform and, where relevant, law enforcement. Keep the original file, your detector report and the reverse search hits as evidence.
FAQ
Can deepfakes be detected with the naked eye?
Often, yes. Low-effort fakes betray themselves at the hands, ears and teeth. For polished fakes, combine human review with detector tools.
Is one AI detector enough?
No. Treat any detector score as a probability. Always pair it with EXIF and reverse image checks.
Do deepfakes always lose EXIF?
Most fully generated images carry no real camera EXIF. Edited deepfakes can retain partial EXIF from the source photo, which is itself a useful clue.
Conclusion
Reliable deepfake detection in 2026 is not about one magic tool. It is a short, repeatable workflow: look, read metadata, run a detector, reverse search, and check the heatmap. PhotoRadar bundles every step so you can finish a check in under five minutes.