A fake photo checker, usually an AI image detector, can tell you whether a picture was likely generated by AI and can often flag big AI edits like face swaps. It can’t tell you whether a real photo has a false caption, whether a scene was staged, or reliably whether someone made a small manual edit. Most “fake photos” that spread online are real photos in a false context, so a checker is one tool in a small kit, not a single answer.
Five kinds of fake photos #
“Fake” covers very different things, and each needs a different check:
| Kind of fake | Example | Does a fake photo checker catch it? | What catches it |
|---|---|---|---|
| Fully AI-generated | A photorealistic image of an event that never happened | Often | AI detector, metadata, source checks |
| AI-edited | A face swap, an object added or removed with generative fill | Sometimes, better for big edits | AI detector on a crop, comparison with the original |
| Manually edited | Cloning, compositing, color changes in a photo editor | Rarely | Comparison with the original, close inspection |
| Real but miscaptioned | A 2017 flood photo shared as today’s | No | Reverse image search, date and location checks |
| Staged | A real photo of a scene set up to mislead | No | Source checks, other angles, reporting |
The last two rows are where most misinformation lives, and an AI checker will correctly say “Likely real” about both. That’s not a flaw in the checker. The picture really is real. The deception is in the story around it.
How a fake photo checker works #
Most checkers combine some of these:
- Metadata and Content Credentials. Some AI tools write their name or a signed C2PA credential into the file. When that survives, it’s strong evidence.
- A trained classifier. A machine learning model that has seen large numbers of real and AI images and learned the statistical fingerprints generators leave in texture and noise.
- Watermark checks. Some companies embed invisible watermarks. Google’s SynthID can be checked in the Gemini app, but only for Google AI content.
What to look for in a fake photo checker #
- It admits uncertainty. Detection is probabilistic. A checker that always gives a confident verdict is hiding its error rate. A good one has a middle “Uncertain” result.
- It shows its reasoning. You should be able to see whether metadata or the model decided, and how confident it is.
- It handles screenshots. Many images you’ll check are screenshots. A crop tool that removes app interface makes a real difference.
- It respects privacy. Online checkers upload your image. For private photos, an on-device checker keeps it on your phone.
- It’s updated. New generators appear regularly, and old models miss them.
A phone-based example: Expose AI #
Expose AI is a fake photo checker for iPhone and Android built on those principles:
- Three verdicts: Likely real, Uncertain or Likely AI-generated, with a confidence meter.
- Two passes on the device: a metadata read for Content Credentials, XMP and generator tags, and a neural network trained on the visual fingerprints of AI generators. The “How we decided” card names the pass that made the call.
- Share from any app, then Crop to isolate the picture inside a screenshot. It spots screenshots on its own and suggests cropping.
- Face swaps and AI edits are within what it targets, as well as fully synthetic images.
- Private: no account, no upload, works offline.
- Free with ads; Pro ($4.99 a month or $29.99 a year) adds a larger on-device model with roughly half as many Uncertain verdicts on the developer’s evaluation set, batch check for up to 12 images, and no ads.
Its honest limits: it checks still images only, not video. Heavily compressed images, screenshots of screenshots and brand-new generators are harder to judge, and metadata can be stripped or faked, so the metadata pass is never the only one. The app itself calls a result a signal, not proof.
A full fake photo check in five minutes #
- Reverse search the image to catch miscaptioned and recycled photos.
- Find the original source and check whether the account is credible.
- Run a fake photo checker for AI generation and AI edits.
- Look closely at text, hands, reflections and shadows.
- Check the context: does the weather, location and timing fit the claim?
For breaking news, how to check if a viral photo is real walks through the routine with examples, and how to verify image authenticity covers the deeper checks professionals use.
How to read the result #
- Likely AI-generated: strong evidence of AI. Confirm with source checks before accusing anyone, especially if the image shows a real person.
- Uncertain: not enough signal. Find a larger copy or rely on other checks.
- Likely real: no AI signals found. The image may still be miscaptioned, staged or manually edited.
We explain scores, confidence and common misreadings in AI image checker results explained.
Frequently asked questions #
Is there a free fake photo checker? #
Yes. Expose AI offers unlimited free checks on your phone with ads, and web tools like Hive Detect are free to use. Some online services, like AI or Not, have free tiers with monthly limits. Our best AI image detector roundup compares them.
Can a fake photo checker detect Photoshop? #
Not reliably. AI checkers are trained to spot generated pixels, so they may catch generative fill or a face swap, but traditional edits like cloning or compositing aren’t what they look for. Comparing with the original image is the best way to catch manual edits.
Can a fake photo checker be wrong? #
Yes, in both directions. Filtered or heavily processed real photos can be flagged as AI, and some AI images slip through, particularly from new generators or after heavy compression. Treat a result as a strong signal and combine it with source and context checks.
How do I check if a photo of a person is fake? #
Check the account posting it, reverse search the photo to see if it’s stolen from someone else, look closely at ears, hair edges, accessories and background, and run it through a checker. If the photo is a profile picture, the account around it often gives the game away faster than the image does.