Detect a Photoshopped Image Online — Free
Run error level analysis on a JPEG to find regions that have been through a different number of compression passes than the rest of the image. Edited areas often light up. Read the result as evidence to investigate, never as proof.
How to do it
- Drop the image you want to check.
- Adjust the amplification so the error-level differences become visible.
- Look for regions that glow differently from their surroundings — those were re-saved separately. Treat this as a hint, not proof.
- Download the analysis view if you need to share what you found.
What error level analysis measures
Every time a JPEG is saved, it acquires a compression signature. Re-compress the whole image and every region degrades by a broadly similar amount. But paste in a region from another file, or paint over an area and save, and that region has now been compressed a different number of times from its surroundings. Error level analysis re-saves the image at a known quality, subtracts the result from the original, and amplifies the difference. Areas with a different compression history stand out as brighter or differently textured than the rest.
How to read the output
You are looking for inconsistency, not brightness. Uniform texture across the whole frame suggests a single compression history. A rectangular patch that is markedly brighter than everything around it, or an object whose edges glow while similar objects do not, is worth investigating. What you are not looking for is a bright area on its own: high-contrast edges, text, and sharp detail are legitimately brighter in every ELA output, because they genuinely carry more compression error. The signal is the mismatch between similar things.
Why this is not proof
ELA produces false positives constantly and false negatives just as easily. Every social platform re-compresses uploads, which flattens the very differences the technique depends on — an image that has been through Facebook is often unreadable by ELA regardless of its history. Screenshots destroy the evidence entirely, because the screenshot is a single fresh compression of everything. A skilled editor who re-saves the whole composite at a uniform quality erases the signature deliberately. Treat a suspicious result as a reason to look harder, never as a conclusion.
What else to check
ELA is one input among several. Look at lighting direction and whether shadows agree across objects. Check whether reflections are consistent. Look for repeated texture patches, the fingerprint of clone-stamping. Examine EXIF metadata, if any survives, for editing software names and timestamp inconsistencies. Reverse image search often settles the question in seconds by finding the unedited original. Any single technique can be defeated; several pointing the same way is considerably harder to fake.
Limits
The amplification level is adjustable, defaulting to 15 — raise it to make faint differences visible, lower it when the whole frame is saturated. Input up to 5 MB. The technique only works meaningfully on JPEGs, since it depends on JPEG compression history. Everything runs in your browser, so an image you are analysing in a dispute is never uploaded to anyone else's server.
Frequently asked questions
Can this prove an image was edited?
No. It produces a signal worth investigating, not a verdict. False positives and false negatives are both common, and a careful editor can erase the signature.
What am I looking for in the result?
Inconsistency between similar things. A rectangular patch brighter than its surroundings, or one object glowing while comparable objects do not. Bright edges alone are normal.
Why does an image from Facebook show nothing useful?
Social platforms re-compress every upload, which flattens the differences the technique depends on. The same is true of screenshots.
Does it work on PNG files?
Not meaningfully. The technique depends on JPEG compression history, and PNG is lossless so there is no such history to analyse.
What should the amplification be set to?
15 is the default. Raise it to bring out faint differences, lower it if the entire frame saturates and detail is lost.
What else should I check?
Shadow and lighting consistency, reflections, repeated texture from clone-stamping, surviving EXIF metadata, and a reverse image search for the original.