How to Check Whether an Image Is AI-Generated: 6 Clues

Updated 5 November 2026 · Standard formula, shown below

Checking whether an image was AI-generated means weighing visible details, file information, and any available source evidence—not relying on one supposed giveaway. You will need the image itself and, if possible, access to its original file and the page or person that shared it. The six clues below can help you judge whether an image needs closer verification, but none proves its origin by itself.

What You’ll Need

The image at the highest resolution you can access. A screenshot or social-media preview may have been resized or compressed.
A browser and, optionally, a reverse-image-search service or image-forensics tool. Treat automated results as leads, not verdicts.
The original file or a link to its earliest known source, if available.
A few minutes to compare details and check the image’s context.

Step-by-Step

Preserve the best available copy. Open the image’s original post or source rather than examining a thumbnail. If you are authorized to download it, keep an untouched copy for inspection. Avoid resaving it through an editor first: that can remove metadata or change compression artifacts. Note where you found it and when, since a caption or repost may not identify the original creator.
Inspect small details at normal size and close up. Look at hands, teeth, ears, jewelry, glasses, and objects being held. Check whether fingers connect naturally, earrings match, and object edges make sense. Then inspect text, logos, repeating patterns, and fine textures. AI image generators can produce misspelled lettering, inconsistent patterns, or details that blur into nearby objects. But ordinary photography, motion blur, low resolution, and editing can cause similar defects. One odd hand is not proof.
Check lighting, reflections, and physical consistency. Compare shadows and highlights across the scene: do they point in plausible directions, and do objects seem to sit on the same surface? Look for reflections that fail to match nearby people or objects, inconsistent perspective, or a background that appears to merge into a subject. Also consider whether the scene could simply involve unusual lighting, a wide-angle lens, a shallow depth of field, or deliberate retouching.
Look for repeated or implausibly smooth visual patterns. Examine hair, foliage, fabric, skin, and backgrounds. Repeated shapes, waxy surfaces, or texture that fades into meaningless detail may raise questions. Zooming too far can make normal pixels look strange, especially after an image has been compressed. Judge whether the pattern is genuinely inconsistent at the image’s intended viewing size, and compare it with similar photographs if possible.
Check the file and any provenance information. If you have the original file, view its properties or use a reputable metadata viewer to see whether it includes a camera model, editing history, or content credentials. Some images may carry signed provenance data that records information about creation or edits. A credible, verifiable record can support a conclusion, but it may be absent, stripped during sharing, or incomplete. Metadata can also be edited, so ordinary fields are not proof. No camera metadata does not mean an image is AI-generated.
Search for earlier copies and corroborate the context. Use reverse image search to find visually similar or earlier versions. Check whether the image appeared before the event it claims to show, whether reputable sources identify its origin, and whether the account sharing it provides a traceable source. A matching image might be an older photograph, a crop, or an edited version. Compare the actual image, not just similar-looking results. If the source cannot be established, describe the image as unverified rather than declaring it AI-generated.

Common Mistakes

Treating a visual glitch as a verdict: AI can create artifacts, but photographers and editing software can too. Require several independent clues before raising confidence.

Trusting detector scores as certainty: AI-image detectors can misclassify both generated and genuine images. Their results vary with image quality, editing, model type, and the tool’s training data. Do not use a score alone to accuse someone or make a consequential decision.

Assuming metadata settles the question: Metadata may be missing, altered, or removed by a platform. Conversely, a file can contain camera details and still have been substantially edited.

Confusing AI editing with full generation: A real photograph may have AI-based objects removed, backgrounds extended, or details altered. Ask whether you are checking the whole image’s origin or whether a particular part has been changed.

Ignoring ordinary image processing: Cropping, sharpening, resizing, filters, and compression can create strange edges and textures. Inspect a better-quality copy before drawing conclusions.

Tips

Separate observations from conclusions: write down the specific detail that looks inconsistent, then consider alternative explanations.
Compare evidence from different sources, such as visual inspection, provenance data, and the image’s publication history. Several versions of the same detector are not necessarily independent evidence.
Keep uncertainty visible. “Possibly AI-generated” or “origin unverified” is more accurate than a definite label when evidence is inconclusive.
If the image matters for reporting, research, or a dispute, preserve the original file and source link, and ask a qualified image-forensics professional to assess it.

Troubleshooting

The image is too small or blurry: Find the original post or a higher-resolution copy. Do not enlarge a tiny preview and treat invented-looking pixels as evidence; upscaling cannot restore details that were never captured.

The file has no metadata: This is common after social platforms process uploads. Continue with visual checks and source research. If possible, ask the uploader for the original file and information about how it was made.

Search finds no earlier copy: That does not establish that the image is genuine or generated. Search a distinctive crop, try another search provider, and check the source’s other posts and cited references. A new or private image may not be indexed.

Clues point in different directions: Record what supports each possibility and avoid forcing a yes-or-no answer. Seek independent review if the distinction has meaningful consequences.

A detector gives a strong result but other evidence disagrees: Treat the score as one uncertain signal. Check the tool’s stated limitations and test a clean, high-quality copy if appropriate; do not upload sensitive images to an unknown service.

Frequently Asked Questions

Can I tell for certain just by looking?
No. Visual clues can suggest closer examination, but image generation and editing improve continually, and ordinary photography can contain similar artifacts.

Are AI-image detectors reliable?
They can offer a useful lead but can produce false positives and false negatives. Accuracy depends on the image and detector, so use other evidence too.

Does missing camera metadata mean an image is AI-generated?
No. Many websites and messaging apps remove metadata, and screenshots often do not preserve the original camera information.

Can a genuine photo have AI-generated parts?
Yes. A photograph can be edited with generative tools. Examine the specific area in question and distinguish AI-assisted editing from an image created entirely from a prompt.

What is the strongest clue?
There is no universal single clue. Verifiable provenance or a trustworthy original source can be especially useful, while visual artifacts and detector scores need corroboration.

What should I say if I cannot confirm the origin?
Say that the image’s origin is unverified and explain what you checked. Avoid presenting an uncertain assessment as fact.

Method: the standard formula for this calculation, rounded as each result states. See how the calculators work or report an error.