Error Level Analysis
ELA re-saves an image at a known quality and maps the difference. Areas with a different compression history stand out — which is why it has been the standard first look at a suspected edit for over a decade. It is also the most over-interpreted tool in image forensics, so the result comes with instructions.
How Error Level Analysis works
The principle is simple, which is both its strength and the reason it gets misused so often.
Every JPEG save loses information
Compression discards detail in a predictable way. Save the same image again and the second save loses less, because much is already gone.
Different histories, different error levels
A region pasted in from another file has been through a different number of saves. Re-save the whole image and that region responds differently.
The map shows the difference
Bright areas returned a larger difference between the original and the re-saved version. It is a map of compression response, not a map of lies.
Edges always light up
Sharp boundaries, text and high-contrast detail produce high error levels naturally. This is the single most common misreading of an ELA image.
Look for regions, not brightness
The signal is an area that differs from **similar** surroundings — smooth sky next to smooth sky, skin next to skin. Bright edges mean nothing.
Uniformity is its own signal
Photographs vary because textures compress differently. An unusually flat map can indicate generation — or simply heavy compression, which is why it is a hint and not a finding.
Reading the result honestly
Compare like with like
Find two areas of similar texture and see whether they respond similarly. Comparing a face against a brick wall tells you nothing.
Rule out the innocent explanations
Resizing, format conversion, repeated saving and platform re-encoding all produce ELA patterns that resemble editing. Most bright regions are not edits.
Treat it as one input
ELA is genuinely useful for locating a suspected edit and genuinely bad as a standalone verdict. In our full analysis it carries a deliberately modest weight alongside eleven other methods.
ELA finds editing, not generation. To check whether an image was made by AI at all, start with our AI image detector — it covers any image type and runs the same twelve methods.
When ELA helps
- Locating a suspected edit — You think something was removed or added, and want to see where.
- Checking a claim of authenticity — An image presented as straight from camera.
- Learning image forensics — The most intuitive way to see compression history.
- Before a deeper analysis — A quick look that tells you whether to dig further.
Frequently asked questions
Related tools
Check your first image now
Create a free account and get 3 checks. Full breakdown across all 12 methods, no card required.