How to Check If a Photo Is AI-Generated (2026)

There are two reliable ways to check whether a photo is AI-generated: read what's *inside the file* (provenance metadata like C2PA), and read the *image itself* (visual tells). The metadata check is the more definitive one when it's present; the visual check is your fallback when the file's been stripped. Use both.

Here's how to actually do each.

Method 1: check the file's metadata

Many AI generators stamp images with provenance data — most notably C2PA Content Credentials, plus EXIF "software" tags. This is the closest thing to a smoking gun, because it's the file declaring its own origin.

How to check:

1. Run the image through a Content Credentials / metadata scanner. A scanner inspects the file for a C2PA manifest, EXIF, and related tags.
2. Look for a C2PA/JUMBF manifest or an AI "software" signature. If it's there, the file is announcing it was AI-made or AI-edited.
3. A clean scan isn't proof of "real." Metadata can be stripped by re-encoding (see removing AI watermarks and metadata), so absence of a tag just means you move to Method 2.

If you want to understand exactly what these manifests are, our explainer on C2PA metadata breaks it down.

Method 2: read the image for visual tells

When metadata is missing, your eyes are the tool. AI images have characteristic weak spots — usually *around* the subject, not the face:

• Hands and fingers — extra, fused, or bent-wrong digits.
• Text — signs, labels, and logos render as garbled pseudo-letters.
• Backgrounds — melted bystanders, warped lines, objects that merge.
• Symmetry slips — mismatched earrings, glasses arms that don't line up, teeth that blur.
• Skin and texture — unnaturally smooth, waxy, or "too perfect."
• Accessories — jewelry, straps, and patterns that don't connect logically.

No single tell is conclusive on its own; look for two or three together.

Tools and their limits

Standalone "AI detector" websites exist, but treat their scores as *hints, not verdicts* — they produce both false positives (flagging real photos) and false negatives (missing AI). The metadata check is more trustworthy when a tag is present; the visual check is more trustworthy than a black-box score you can't inspect. Combine the evidence rather than trusting one number.

Doing this on dating apps

If you're trying to tell whether a *match's* photos are AI, the same two methods apply, with a dating-specific lens — we cover the practical version in how to tell if someone's photos are AI. And if you're checking your *own* AI photos before posting, the goal is the reverse: confirm the file is clean and the image has no obvious tells.

Frequently Asked Questions

What's the most reliable way to check? If the file still has its metadata, a C2PA/Content Credentials scan is the strongest signal. If it's been stripped, fall back to visual tells like hands, text, and backgrounds.

Can AI images be undetectable? There's no guarantee either way — detection is an arms race. A stripped, well-made image with no obvious tells can be very hard to call, which is exactly why no honest tool promises 100% accuracy.

Do AI detector websites work? Somewhat, but they're inconsistent — expect false positives and negatives. Use them as one input alongside metadata and visual inspection, not as the final word.

How do I make sure my own photos look clean? Curate out any frame with visual tells, and use files with no AI metadata. Our [generated photos](https://www.matchmaxing.com) ship with metadata stripped, but you should still curate for believability.