keepcalculator

06 Numbers and Powers

AI-Generated Images vs Edited Photos: How to Tell

Card No. 06·84418 · Updated · Standard formula, shown below

To tell an AI-generated image from an edited photo, look for several clues together: inconsistent details, implausible lighting, image history, and reliable provenance information. No single visual test is conclusive—generative tools can make convincing pictures, while ordinary editing can create artifacts that look synthetic.

What counts as generated or edited?

An AI-generated image is made wholly or partly by a generative model, often from a text prompt or another image. An edited photo begins with a camera capture and is altered—for example, by cropping, retouching, replacing a background, or adjusting color. The boundary can blur: a photographer might use generative fill to extend a real photograph, or combine a real subject with an AI-created setting.

It helps to distinguish how an image was made from whether it is truthful. A genuine photograph can be misleadingly cropped or manipulated, and a clearly labeled AI illustration can be harmless and accurate for its purpose. The practical question is usually whether the image represents a real scene or event as presented.

Visual clues to inspect

Inspect the whole picture first, then zoom into areas where objects meet or fine detail matters. Look for contradictions rather than treating one odd feature as proof.

  • Hands, faces, and repeated details: Check fingers, ears, teeth, jewelry, buttons, and patterns. Unusual anatomy or details that change from one side to another can be a clue, though current image generators often handle these well.
  • Text and symbols: Letters on signs, labels, clothing, or screens may be garbled, inconsistent, or meaningless. Real photos can also contain motion blur, glare, and low-resolution text, so context matters.
  • Geometry and object relationships: Inspect glasses, rails, furniture, reflections, and objects partially hidden behind others. Lines may bend oddly, or an object may merge into its surroundings.
  • Lighting and shadows: Ask whether highlights and shadows agree with the apparent light sources. A face, nearby wall, and cast shadow lit in incompatible directions deserve closer examination.
  • Fine textures: Hair, foliage, patterned fabric, and crowds may look smeared, repetitive, or detailed in a way that dissolves on close inspection. Compression and sharpening can create similar effects.

These clues are strongest when multiple unrelated inconsistencies occur in a high-quality image. A single strange hand or shadow is weak evidence: lenses, reflections, perspective, and image compression can make real scenes look peculiar.

Check the image’s history and source

Provenance can provide evidence that visual inspection cannot. Find the earliest available version, identify who published it, and check whether a credible photographer, news organization, institution, or original source provides context. Search for the image or distinctive parts of it to see whether an older version appears with a different caption. Reverse image search can uncover reuse, but it cannot by itself establish whether an image is genuine.

Some files carry metadata, such as camera model, date, or editing software. Metadata is useful context, not proof: it can be stripped during sharing, changed, or absent from images exported by ordinary editing tools. A camera name in metadata does not certify that the pictured scene is unaltered.

Content credentials and other provenance records may document a file’s origin and edits when they are created and preserved. Their presence can support a chain of custody, but they do not automatically prove that a depicted event happened as claimed. Their absence proves little, because not every camera or service creates or retains such records.

Use detection tools cautiously

AI-image detectors estimate whether an image resembles examples produced by generative systems. Their results are not definitive authentication. Performance can vary with the model, image type, resolution, compression, cropping, and later edits. A detector may flag a heavily edited photograph or miss a generated image, so treat a score as a prompt for further checking—not a verdict.

Likewise, forensic signs such as inconsistent noise, unusual edge patterns, or mismatched compression can indicate editing, but they are not exclusive to AI generation. Resizing, screenshots, social-media processing, and repeated saving can alter these signals. If an image matters—for example, as evidence in a report—seek the original file and independent corroboration rather than relying on a screenshot or automated score.

A practical verification process

  1. Clarify the claim. Write down what the image is said to show, where, and when. An authentic image can be paired with a false date, location, or caption.
  2. Find the source. Trace it to the earliest available publisher and look for an original post, photographer, or related reporting.
  3. Inspect specific details. Check text, reflections, shadows, anatomy, object boundaries, and repeated patterns at normal size and when enlarged.
  4. Compare independent evidence. Look for other photographs, video, eyewitness accounts, or reliable records that corroborate the same event. Similar copies of one image are not independent confirmation.
  5. Check provenance and metadata. Treat any records as supporting clues, and note whether the file may have passed through services that remove information.
  6. Keep the conclusion calibrated. Say “unverified” or “possibly altered” when evidence is incomplete. Avoid calling an image fake solely because a detector or one visual oddity suggests it.

Common misconceptions and limits

“It looks realistic, so it must be a photo” is unreliable: convincing synthetic images can look natural at a glance. The reverse claim is also mistaken. A blurry face, strange shadow, or imperfect hand does not prove generation; cameras capture motion, distortion, and awkward moments, while editing can introduce artifacts.

There is no universal checklist that settles every case. Generative systems change, images are often compressed or edited after creation, and authentic photographs may be composites or retouched. The strongest assessment combines visual inspection, source research, provenance when available, and corroboration. When those sources disagree or are missing, the honest answer may be that the image’s origin cannot be determined confidently.

Frequently Asked Questions

Can you tell if an image is AI-generated just by looking?
Sometimes visible inconsistencies provide useful clues, but visual inspection alone cannot reliably classify every image. Check multiple details and the image’s source.

Are AI image detectors accurate?
They can offer a probabilistic signal, but results may be wrong and can change with compression, editing, or image type. Do not treat a detector result as proof.

Does missing photo metadata mean an image is fake?
No. Websites and messaging services often remove metadata, and many legitimate images lack useful records. Missing metadata does not establish how an image was made.

Can an edited real photo look AI-generated?
Yes. Retouching, compositing, filters, resizing, and compression can create odd textures or edges that resemble generation artifacts.

What is the strongest way to verify a viral image?
Trace it to the earliest credible source, check its context and available provenance, and seek independent evidence for the event it claims to show. No single clue is enough in every case.

Related

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

More cards in this drawer

The whole Numbers and Powers drawer →