How to Tell If a Photo Has Been Edited or Photoshopped: 5 Free Checks and Their Limits

Last reviewed October 5, 2026 · 9 min read · By the PEA team (Lumina Spark)

Short answer: There are five free checks: look for an earlier version with a reverse image search, read the metadata, check for Content Credentials, run error level analysis, and look for physical inconsistencies. Each can turn up signs of editing, but none can prove a photo is unedited. After a photo has passed through social media, a messaging app or a screenshot, the honest result is often "not enough to judge."

First, decide what kind of edit you mean

"Edited" covers two different things:

  • Adjustments change how a photo looks, not what it shows: brightness, color, cropping, resizing. They rarely matter in a dispute, unless a crop cuts out something important.
  • Content changes alter what the photo shows: pasting in part of another image, removing a person or object, warping a body or background, or using AI to erase, add or replace part of the scene.

Content changes are now built into everyday apps: Magic Eraser and Magic Editor in Google Photos, Clean Up and Extend on recent iPhones with Apple Intelligence, and Generative Fill in Photoshop, which rewrites a selected area from a text prompt or the surrounding pixels. So "edited with AI" usually means a partial edit like these. For wholly AI-made images, see checking whether a photo is AI-generated.

Before you start, get the original file from the person who took it, not a screenshot or forwarded copy, and work on copies.

The five checks at a glance

CheckWhat it can showLimits
1. Reverse image searchAn earlier or different version online, or earlier use in another contextNo match doesn't mean the photo is original; new personal photos aren't indexed
2. Metadata (EXIF)Device, date and time, camera settings, sometimes editing software or an AI-edit labelOften stripped by social media and messaging apps; easy to change
3. Content Credentials (C2PA)A signed record of how the file was made and edited, and whether it changed after signingMost images don't carry them yet; often lost on upload or screenshot
4. Error level analysis (ELA)Areas with a different compression history, which can point to pasted or retouched regionsEasy to misread; repeated saving and recompression wipe out the signal
5. Visual inconsistenciesShadows, reflections, bent lines, edges and repeated textures that don't make physical senseSkilled edits and AI fills may leave nothing visible

Use them together. One check pointing to editing is a lead; several agreeing are stronger. Finding nothing proves nothing.

Check 1: Look for an earlier version (reverse image search)

If the image has been online before, a reverse image search may find the original, and comparing the two is often the fastest way to see what changed.

  • Google Lens: on a computer, click "Search by image" on google.com, or right-click an image in Chrome and choose "Search with Google Lens." On a phone, use Lens in the Google app (or Chrome on iPhone). Where available, "About this image" shows when Google may have first seen a similar version.
  • TinEye: sort matches to find the earliest copy TinEye knows of or the most changed version, and switch between your image and a match to highlight cropping or other changes. TinEye says search images are never saved or indexed.

Look for missing or added objects, changed text, a different background, or the same picture published years earlier about another event. But no match proves nothing: a photo taken last week on someone's phone usually won't be in any index.

Check 2: Read the metadata (EXIF)

Photos straight from a camera or phone usually carry EXIF metadata: device, date and time, lens and exposure settings, sometimes location. Some editing apps also write their name into the file.

  • Windows: right-click the file, choose Properties, then the Details tab.
  • Mac: open the file in Preview and choose Tools > Show Inspector.
  • iPhone: in Photos, open the photo and swipe up or tap the Info button. Depending on the photo, Apple says you'll see the device, the date and time, and camera details such as lens and shutter speed.

Some apps label AI edits: Google says photos edited with Magic Editor, Magic Eraser or Zoom Enhance in Google Photos carry IPTC metadata saying so, shown with the photo's details in the app.

The limits:

  • It can be changed. Apple's own Photos app lets you adjust a photo's date, time and location. NIST notes that anyone copying or editing a file can falsify its metadata, and Google warns that credit and source-type labels can be modified with editing tools.
  • It's often stripped when files are shared, for example on social media. So missing metadata is not evidence of editing.

Treat metadata as consistent or inconsistent with the story you've been told, not as proof.

Check 3: Check for Content Credentials (C2PA)

Content Credentials, based on an open standard from the Coalition for Content Provenance and Authenticity (C2PA), let a camera, phone, editing app or AI tool attach a signed record of how a file was created and edited. The Content Authenticity Initiative's free Verify tool (contentcredentials.org/verify) accepts JPEG, PNG, HEIC, WebP and other formats and shows any credentials it finds.

  • Credentials present and valid: strong context. Any change to the file or its record after signing, however small, breaks the match. But the record shows what the signer asserted, and C2PA itself says provenance alone can't tell you whether content is true.
  • No credentials: this tells you very little. Verify notes that Content Credentials are still rolling out, and C2PA confirms they can be removed; uploads, screenshots and format changes often strip them.

Check 4: Run error level analysis (ELA)

When part of a JPEG is pasted in or retouched and the file is saved again, that area can have a different compression history from the rest. Error level analysis, free on FotoForensics, re-saves the image at a known quality and shows how much each area changes.

ELA is easy to misread:

  • Compare like with like. Flat areas such as a clear sky naturally look darker than high-contrast edges, so FotoForensics' tutorial says to compare edges with edges and surfaces with surfaces.
  • Small edits can vanish. A single-pixel change or minor color adjustment may not show, and an image resaved many times can end up at a uniform minimum error level.
  • It's raw data, not a verdict. The site says it doesn't draw conclusions.

Mind your privacy: FotoForensics calls itself a public site with no privacy, keeps pictures for at least three months, possibly indefinitely, and uses uploads for research. Don't upload sensitive photos.

Check 5: Look for physical inconsistencies

Open the largest version you have and zoom in:

  • Light and shadows: do shadows fall the same way and match the light source?
  • Reflections: do mirrors, windows and water show what should be there?
  • Straight lines: door frames, tiles or a horizon that bend near a person's body suggest warping.
  • Edges: halos, blur or unnaturally sharp outlines around a person or object.
  • Repeated textures: identical patches of grass or brick can mean something was cloned out.
  • Text: signs or documents with mismatched fonts, spacing or angles.

These clues prompt good questions, but well-made edits, especially AI fills, may leave nothing visible.

Choosing free tools, online or on an iPhone

Most sites and apps that promise to tell you whether a photo is photoshopped do one of the checks above, so know which one you're using. On an iPhone you can cover three without a computer: the Info panel in Photos, Google Lens in the Google app, and the Verify website in your browser.

Be wary of a single score that doesn't explain what was found, and before uploading a private photo, check where it goes, how long it's kept and who can see it.

AI erase and generative fill: why they're harder to spot

AI erase tools fill the gap with content generated to match the surroundings, so there's often no obvious seam. Sometimes the filled area has different grain or noise, detail that turns mushy or oddly smooth, or light that doesn't quite match. ELA, close inspection and any AI-edit labels in the metadata or Content Credentials are your main options.

Be cautious with automated AI detectors. NIST's 2024 report on synthetic content found that detectors tend to show substantial error rates on images from generators they weren't trained on, particularly after compression and resizing. Researchers have also found that biases in common training datasets can lead detectors to pick up on JPEG compression and image size rather than real signs of generation.

When you can't tell: social media, messaging apps and screenshots

Photos posted to social media, sent through messaging apps or captured as screenshots are usually resized and recompressed, often losing their metadata and Content Credentials. Recompression overwrites the fine traces ELA relies on, and a screenshot has no camera metadata at all.

A degraded genuine photo and an edited one whose traces have washed out can then look the same. If your checks find nothing, the accurate conclusion is "not enough to judge," not "unedited." Ask for the original file, straight from the device that took it, and check again. For screenshots of messages and posts, see how to authenticate text message screenshots for court.

If the photo matters in a dispute

If the image may become evidence, with a business, an insurer, an employer or in court, do this first:

  1. Keep the file exactly as received. Don't re-save or re-screenshot it. Work on copies.
  2. Record where it came from: who sent it, when, and through which app or account.
  3. Hash it. A SHA-256 hash lets you show later that the file hasn't changed by a single byte. On Windows, PowerShell's Get-FileHash uses SHA-256 by default; on a Mac, run shasum -a 256 followed by the file name in Terminal.
  4. Log every check: the tool, the date and the result in the tool's own words, with a screenshot.
  5. Get professional help when the stakes are high. Whether evidence is admitted, and how much weight it gets, is for the court. A lawyer and, where justified, a forensic examiner can go further than any free tool.

What PEA can and can't do

PEA (Professional Evidence Authenticator), from Lumina Spark in Japan, runs several independent analyses on an uploaded image and shows its findings, separating confirmed facts from concerns, with a grade from 1 to 10. It checks Content Credentials (C2PA) and, where available, some official watermark checks. When there isn't enough to go on (a frame from a video, a photo of a screen or printout, or no usable material), it says "Not assessed (Grade –)" instead of guessing, and suggests a human review. Its certificate fixes the file with a SHA-256 hash and seals it with an Ed25519 digital signature and an RFC 3161 timestamp from a third-party time-stamping authority, which anyone can verify by certificate number on PEA's public verification page. It does not prove when or by whom a photo was taken. Staff see an image only if you request a human review and consent, and free results are deleted after 7 days.

PEA runs in Japanese today and the English version is in early access. If it sounds useful, you're welcome to join the early-access list at our early-access page.

This article is general information, not legal advice.

Sources