A real photo gets flagged as AI when it shares the features detectors associate with generated images: skin smoothed by a beauty filter, heavy phone processing, an AI edit or upscale somewhere in the image, severe compression, or the flawless lighting of a studio shot. Detectors work on probabilities, and those features push the score toward “AI” even when a camera took the picture. The fix is to check the original, unfiltered file, and if you need to prove a photo is real, to show the evidence around it: originals, other shots from the same moment, and metadata.
The usual causes #
| Cause | Why it looks like AI |
|---|---|
| Beauty filters and face smoothing | Removes pores and texture, producing the same waxy skin generators make |
| AI enhancement and upscaling | Rebuilds fine detail with a generative model, so some pixels really are AI |
| Object removal and generative fill | Adds a generated patch to a real photo |
| Heavy phone processing (night mode, HDR, portrait mode) | Merges frames and smooths noise, erasing natural camera grain |
| Compression and repeated reposts | Destroys the fine texture that distinguishes camera noise |
| Screenshots | Adds interface elements and resampling |
| Studio or product photography | Perfect lighting and clean backgrounds resemble polished AI output |
| 3D renders and digital art | Not photos at all, and not always AI either, but unlike camera images |
Filters and AI touch-ups #
This is the most common cause. Many phone cameras and social apps apply face smoothing by default, and editing apps offer AI enhancement with one tap. Those features use AI models, so in a technical sense the photo has been partly processed by AI. A detector isn’t entirely wrong to notice.
Google Photos even records this: it marks photos edited with Google AI tools like Magic Editor, Magic Eraser and Zoom Enhance in their metadata (Google). If you used one of those tools, a checker that reads metadata may see that note.
Computational photography #
Modern phones rarely take a single exposure. Night mode, HDR and similar features merge several frames and clean up the result. The IPTC even added a separate metadata value for this, “computationalCapture”, for multi-frame captures merged with non-generative processing (IPTC). The output is sharper and cleaner than a raw camera frame, and cleanliness is one of the things detectors notice.
Compression #
A photo that has been sent through a messaging app, posted, screenshotted and reposted has lost most of its fine grain. Without that texture, a detector has less to go on and is more likely to land in the middle or tip the wrong way.
Polished photography #
Professional headshots, product photos on white backgrounds and well-lit food shots can look generated to a person, and sometimes to a model too. They share the even lighting and clean surfaces that generators favor.
How to check whether the flag is right #
- Find the original file. Check the photo as it came off the camera, not a copy from social media.
- Turn filters off and reshoot if you can, then check the unfiltered version.
- Look at the metadata. A Google AI edit note or Content Credentials may explain the flag. See how to check image metadata for AI.
- Check the confidence, not only the label. A marginal score is much weaker evidence than a near-certain one.
- Try a second checker. Detectors trained on different data often disagree on borderline images.
On a phone, Expose AI shows its reasoning, which helps with this kind of troubleshooting. The “How we decided” card lists what the metadata pass found (camera and editor details appear as evidence, never as proof) and whether the on-device model made the call, with a confidence meter. When the signal is weak it says Uncertain rather than forcing a guess. The app itself asks you to treat a result as one tool among several, never as the final word about a picture or a person.
How to prove a photo is real #
If someone has accused you of posting an AI image (a contest entry, a dating profile, a listing, a news photo), a detector result won’t settle it either way. Evidence around the photo will:
- The original file with its full metadata, not a re-saved copy.
- Other shots from the same moment. Phones often capture bursts, and real events produce near-duplicates from slightly different angles. Generated images rarely come with a set of consistent siblings.
- RAW files if your camera shot them.
- Content Credentials at capture. A growing number of cameras and apps can sign photos with C2PA credentials when they’re taken. A valid signature from a camera maker is strong evidence of capture. See what C2PA and Content Credentials are.
- Witnesses and context. Other people’s photos of the same event, a location you can show, a timestamp that matches.
For artists accused of using AI, work-in-progress files, layered project files and process recordings do the same job.
If a platform flagged your photo #
Instagram, Facebook and Threads show an “AI info” label when they detect industry-standard AI signals in a file or when the uploader discloses it. Since September 2024, Meta has moved that label into the post’s menu for content that was only edited with AI, rather than generated (Meta). A label on your post usually means your editing software wrote an AI marker into the file. We cover it in what “AI info” means on Instagram.
Frequently asked questions #
Can a detector be wrong about a real photo? #
Yes. Every AI image detector makes false positives, and filtered, processed or heavily compressed photos are the usual victims. That’s why a flag should lead to more checking rather than a conclusion. See how accurate AI image detectors are.
Does iPhone or Pixel processing make photos look AI? #
It can contribute. Night mode, HDR and portrait mode merge frames and smooth noise, which removes some natural camera texture. Most processed phone photos still read as real, but combined with a beauty filter or heavy compression, they’re more likely to be flagged.
Why was my selfie flagged as AI? #
Selfies are often smoothed by the front camera’s beauty mode or a social app filter, which produces skin that looks like a generator made it. Turn off beauty features, shoot again, and check the unfiltered original.
How do I stop my photos being flagged as AI? #
Share originals rather than filtered or upscaled versions, avoid AI enhancement tools when authenticity matters, and keep your original files. If your camera or app supports Content Credentials, turning them on adds a signed record of capture.