keepcalculator

06 Numbers and Powers

How to Spot AI-Generated Images and Videos

Card No. 06·84422 · Updated · Standard formula, shown below

AI-generated images and videos can look convincing, and a single visual clue rarely proves how something was made. This guide gives you a practical way to inspect a social media post, check its source and context, and decide what you can responsibly conclude before sharing it.

What You’ll Need

  • The post or original image/video, if available, plus its caption, account name, date and link.
  • A web browser and access to reverse-image search or a search engine. You can use a phone or computer.
  • For video, the ability to pause, replay and inspect individual frames. Optional: a trusted fact-checking site or image-forensics tool.

Step-by-Step

  1. Pause before judging or sharing. Treat the post as unverified, especially if it makes a dramatic claim, urges immediate action or appears during a breaking event. Decide what you are checking: whether the image is synthetic, whether it has been edited, or whether the caption describes the scene accurately. Those are different questions. A genuine photo can be misleadingly captioned, and a synthetic image can illustrate a true subject without documenting a real event.
  2. Inspect the post and account. Read the full caption and look for a source, date, location, original creator or disclosure that the media is generated or altered. Check the account’s posting history and whether it links to a credible source. A new account, inconsistent details or a sensational caption are reasons to investigate, not proof of fabrication. Also check whether the platform labels the post; labels and disclosure systems vary, and the absence of a label does not establish that media is authentic.
  3. Look closely at the image. Zoom in, but do not rely on one strange detail. Check whether hands, teeth, ears, jewelry, text, logos, reflections, shadows and repeated patterns remain coherent. Look at the edges of objects and whether people or objects appear to merge, change shape or have implausible proportions. Generated media may contain errors, but ordinary compression, filters, motion blur, low resolution and editing can create similar artifacts. Modern generation tools can also produce convincing details, so a clean-looking image proves little.
  4. Review video across several frames. Pause at different points, especially during turns, fast movement, changes in lighting and interactions between people and objects. Watch for faces or clothing that shift between frames, unnatural blinking or mouth movement, inconsistent shadows, warped backgrounds, objects that appear or disappear, and motion that does not match the scene. Replay the audio separately: check whether speech timing, room sound and visible mouth movements seem consistent. Dubbing, subtitles, editing and poor encoding can explain mismatches, so consider them before drawing a conclusion.
  5. Check where the media came from. Use a reverse-image search with the full image, or crop a distinctive part and search again. For video, take screenshots of several clear, representative frames and search them individually. Look for an earlier upload, a longer original clip, or a credible report that identifies the scene. Search a distinctive phrase from the caption alongside the claimed location or event. A matching result may reveal older footage with a new caption; no result is not proof that the content is new or synthetic.
  6. Verify the claim independently. Find reporting or primary material from sources that are not simply reposting the same clip. For a claimed event, compare the date, location, weather, signs, landmarks and clothing with reliable records or other images. Check whether reputable fact-checkers have examined the specific post. Trace sources back as far as possible: many accounts repeating the same upload are not independent confirmation.
  7. Check available provenance, then weigh all the evidence. Some files or platforms may provide creator disclosures or content credentials describing how media was captured or edited. Check what those details actually say and whether they refer to the file you are viewing. Screenshots and re-uploads may strip metadata, and metadata can be incomplete or altered; it is not a standalone guarantee. Combine provenance, source history, visual inspection and independent verification. If the evidence remains uncertain, describe it as unverified rather than declaring it AI-generated or authentic.
  8. Choose a careful next step. Do not repost a doubtful clip as fact. If you share it for discussion, say what is known and what remains uncertain, and link to the original or a reliable verification. Use the platform’s reporting tools if a post appears deceptive or harmful. Avoid publicly accusing a named person based only on visual clues; preserve the post link and relevant details if you need to report it.

Common Mistakes

  • Treating a visual glitch as proof. Compression, filters, retouching and ordinary camera artifacts can resemble generation errors.
  • Assuming realism means authenticity. A polished image or smooth video can be synthetic, edited or taken out of context.
  • Trusting detection tools as verdicts. Automated detectors can produce false positives and false negatives, especially on compressed, cropped or edited media. Treat results as a clue, not a finding.
  • Confusing “AI-generated” with “false.” The media’s origin and the truth of its caption are separate questions. An AI illustration might accompany an accurate explanation; a real clip might be falsely described.
  • Taking missing metadata as evidence. Social platforms often remove file details during upload, so missing metadata says little about origin.
  • Relying on repetition. Many reposts may all trace to one unsupported source.

Tips

  • Save the original post URL and note the account and date before it is deleted or edited.
  • Search a distinctive frame or image crop, not just the caption. Try more than one frame if the first search is inconclusive.
  • Separate observations from conclusions: “the lettering changes between frames” is more precise than “this is definitely AI.”
  • Check whether a platform label identifies AI use, editing or altered content; these terms are not always interchangeable.
  • When accuracy matters, wait for independent reporting or a statement from a source directly involved. Being uncertain is a valid result.

Troubleshooting

  • Reverse search finds nothing: Crop to a distinctive face, object, sign or landmark; try a different frame and search relevant caption phrases. A new, private or heavily edited post may not be indexed.
  • The video is too blurry to inspect: Find the earliest or highest-quality upload, if possible. Avoid sharpening tools as evidence; they can invent or exaggerate details.
  • Clues point in different directions: Record what supports and weakens each explanation. Seek independent sources, and leave the result unresolved if evidence is insufficient.
  • A detector labels the file synthetic: Check the tool’s stated limits and test the exact file, not a screenshot when the tool expects an original. Compare with source and context evidence rather than treating the score as certainty.
  • The original has disappeared: Look for archived reporting or contemporaneous references, but distinguish a copy from the original and be cautious about unverifiable reposts.

Frequently Asked Questions

Can you identify AI-generated media from one visual clue?

Usually not reliably. A flaw may come from editing or compression, while generated media may have no obvious flaw. Use several checks and verify the source.

Are AI image detectors accurate?

They can help prioritize a closer review, but results are not definitive. Performance varies by tool and by how the image has been edited, resized or compressed.

Does a platform label prove a post is fully AI-generated?

Not necessarily. A label may indicate that some content was generated or altered, depending on the platform’s policy. Read the label’s explanation and assess the specific media.

Can a real video be misleading?

Yes. It may be old, edited, staged or paired with an inaccurate caption. Verify the date, place and claim separately from whether the footage itself is genuine.

What should I do if I cannot tell?

Call it unverified, avoid sharing it as fact, and look for stronger evidence later. Do not make a confident claim from an uncertain result.

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 →