How can I check if a photo is AI-generated?

Read the image's provenance record. Upload the image above. If it carries C2PA Content Credentials, you see which tool created it, including the AI markers from DALL·E, Adobe Firefly and licensed Midjourney integrations. If it has no credentials, no tool can give you a reliable answer. That includes pixel-based AI detectors.

Can AI image detectors be wrong?

Yes, often and in both directions. Pixel-based detectors flag real photos as AI, especially edited or scanned images, and they let many AI images pass as real. Their accuracy also drops as new generator models are released, because they were trained on the artifacts of older ones.

Is this an AI detector?

No. This tool reads the C2PA Content Credentials embedded in an image, a signed record written by the tool that made or edited it. It reports whether credentials exist, who signed them, whether AI generation was recorded, and whether the file changed after signing.

The image has no Content Credentials. Does that mean it is AI?

No. Most images on the internet have no credentials. Either the camera or tool did not attach any, or a platform stripped the metadata on upload. Missing credentials mean the origin is unverified. The image may well be a real photo.

How do I prove my own photo is not AI-generated?

Verify it with the original RAW file. Lumethic checks whether the RAW's sensor noise, bit depth and metadata fit a real camera, then compares the RAW against your exported JPEG. The report lists the checks that ran and their results, and it names what it cannot show. It assumes that nobody can yet fabricate a sensor-consistent RAW, and we revisit that assumption as generators change. No method offers mathematical proof. The RAW comparison is a much stronger check than any detector score. It is free to try and needs no account.

Does this tool upload my image anywhere?

No. The analysis runs entirely in your browser using the open-source C2PA WebAssembly SDK. Your image never leaves your device. Only the optional quick provenance check further down uploads the photo to our servers for analysis.

How do I tell if a photo is AI-generated in 2026?

Read its Content Credentials with this tool first, then run the quick provenance check below to screen its metadata and compression history. After that, look for visual artifacts such as malformed hands or garbled text. Visual signs stopped being conclusive as generators improved. Treat them as a reason to look closer.

Can this tool check for the SynthID watermark?

No. SynthID is an invisible watermark in the pixels, and only Google's and OpenAI's own services detect it, for example the Gemini app or OpenAI's Verify tool. This checker reads the other mark that ChatGPT and the OpenAI API attach to generated images, the C2PA manifest. If that record is intact, it names the generator. No probability score is involved.

Can I check a screenshot or a PNG?

Yes. The browser check reads JPEG, PNG, WebP, AVIF, HEIC, TIFF and SVG files as well as DNG and ARW raw files. The quick provenance check accepts JPEG, PNG, WebP, HEIC/HEIF, AVIF and GIF. Screenshots never carry the original's Content Credentials, because a screenshot is a new image without the record. For screenshots, the quick check usually tells you more.

Is this AI image checker free?

Yes. The browser check is free and unlimited. The quick provenance check is free as well and needs no account.

Free tool

Check If a Photo Is AI-Generated

Upload an image and we read its Content Credentials, the signed record that tools like ChatGPT, DALL·E, Adobe Firefly and licensed Midjourney workflows attach to generated images. The check is free and runs in your browser. Your file is not uploaded.

Step 1. Instant read in your browser

Drag & drop a photo here, or click to select

JPEG, PNG, WebP, AVIF, HEIC, TIFF, SVG, DNG and ARW are supported

Choose a photo

Your photo is analyzed in your browser and never uploaded.

Step 2. Deeper check on our servers

Free quick provenance check

Most images carry no Content Credentials, and that alone proves nothing. The quick check goes further than the browser can. It checks a signer against the C2PA trust list and screens the file's metadata and compression history for editing traces. You get a report you can share. Without credentials it says Unrated and lists what was found.

Your photo goes in once, in Step 1. After the browser has read it, one click sends the same file on to this check.

JPEG, PNG, WebP, HEIC, AVIF and GIF are supported. Free, no account needed.

What this checker can and cannot tell you

What it can tell you

Whether the image carries C2PA Content Credentials, and if so, which tool created or edited it, including recorded AI generation.
Whether the credentials are intact or the file was modified after signing.
Which camera or software produced the credential, when, and what edits were recorded.

What no tool can tell you

If an image has no credentials, no tool can prove whether it is real or AI-generated. Pixel-based AI detectors guess from statistical patterns. They regularly flag real photos as AI and let AI images pass as real.

Every tool has this limit. Detection accuracy drops with each new generator release, and a cryptographic record stays checkable.

Read why AI detectors produce false positives

Need to show that your own photo is not AI-generated?

Photographers who are asked for evidence, for example by a contest or a client, do not need a detector. Verify your camera's RAW file against the exported JPEG and you get a verification report you can share.

Verify your photo for free View a sample report

How the check works

Drop in an image

JPEG, PNG, WebP, AVIF, HEIC, TIFF, SVG, DNG or ARW. The file is analyzed locally in your browser with the open-source C2PA SDK. It never leaves your device.

We read the credentials

The checker reads the C2PA manifest. It shows who created the image and with which tool, whether AI generation was recorded, and whether the signature still matches the pixels.

You see the result

You see what the image's provenance record says, and a clear note when there is no record to read. There is no probability score.

Visual signs of an AI-generated image

Looking closely still helps. These artifacts are worth knowing:

Hands and fingers: extra, missing or fused fingers, impossible joints
Text and signage: garbled letters, invented alphabets, inconsistent fonts
Textures: waxy skin, hair strands that merge or dissolve, repeating patterns
Light and shadow: shadows pointing different directions, reflections that don't match the scene
Backgrounds: warped geometry, melted objects, crowds of distorted faces

Each new generator release removes some of these flaws, and current models often pass a casual look. A visual sign is a reason to read the provenance record. It is not an answer on its own.

Read the full guide on how to tell if a photo is AI-generated

Worked example: the viral World Cup photos, real or AI, checked one by one

Try it yourself: can you spot the generated photo by eye?

AI detector vs. provenance check

	Pixel-based AI detector	Provenance check (this tool)
What it examines	Statistical patterns in the pixels	The cryptographically signed record attached at creation
Output	A probability score ("87% likely AI")	Who signed, with which tool, and whether AI generation was recorded
False positives	Common. Edited or scanned real photos get flagged	None. It reports whether a record exists and what it says
Ages over time	Degrades with every new generator release	Stays verifiable as long as the signature does
Limit	Can be wrong while reporting high confidence	Most images have no record, and a missing record proves nothing

What it examines

Pixel-based AI detector

Statistical patterns in the pixels

Provenance check (this tool)

The cryptographically signed record attached at creation

Output

Pixel-based AI detector

A probability score ("87% likely AI")

Provenance check (this tool)

Who signed, with which tool, and whether AI generation was recorded

False positives

Pixel-based AI detector

Common. Edited or scanned real photos get flagged

Provenance check (this tool)

None. It reports whether a record exists and what it says

Ages over time

Pixel-based AI detector

Degrades with every new generator release

Provenance check (this tool)

Stays verifiable as long as the signature does

Limit

Pixel-based AI detector

Can be wrong while reporting high confidence

Provenance check (this tool)

Most images have no record, and a missing record proves nothing

Why this checker does not guess

Most AI image detectors run a classifier over the pixels and output a confidence score. Independent tests show these classifiers err in both directions. Heavily edited or scanned real photos get flagged as AI, and images from current generators often pass. The percentage attached to the verdict does not tell you which case you are looking at.

The C2PA standard records provenance when the image is created and signs the record. The large AI generators now attach Content Credentials to their output, and several camera makers sign images in the camera. Reading that record tells you what the signing tool declared. See which cameras support C2PA Content Credentials

How provenance differs from detection

Detection tries to judge from the pixels whether an image looks generated, and that gets harder with every model generation. Provenance asks a different question, whether there is a signed record of where the image came from. That question has a yes or no answer. The transparency rules of the EU AI Act and the rules of the large photo contests build on provenance for the same reason.

Read the full comparison of provenance and AI detection

Frequently asked questions

This checker reads C2PA Content Credentials. Missing credentials do not mean an image is AI-generated. Present credentials show what the signing tool recorded. The analysis runs in your browser.