Free photo forensics online.

Photo forensics is a set of pixel-level techniques that surface regions of an image which may have been edited, spliced, or generated. The most common tools are Error Level Analysis (ELA), noise residual maps, and luminance-gradient inspection. None of them prove tampering on their own; they highlight suspicious regions a human analyst can investigate.

ELA re-saves the photo as a JPEG at a known quality and compares the result to the original, then amplifies the per-pixel difference. Regions that have been pasted in, content-aware-filled, or edited tend to compress differently from the surrounding background, so they appear as bright speckles. Whole-image brightness is normal and not a manipulation hint. ELA only works on JPEGs; PNG/WebP/HEIC inputs make the test meaningless.

The noise map subtracts a small-radius blur from the image, leaving only the high-frequency noise. Real camera sensors produce a fairly uniform noise pattern across the frame. Splices and AI inpainting often leave patches with noticeably smoother or rougher texture than their surroundings. Look for sharp boundaries between noise textures rather than overall brightness.

It runs a Sobel edge detector on the photo's luminance channel and shows the gradient magnitude. Lighting direction in a real scene is consistent: shadows fall the same way, highlights line up. Spliced subjects often disagree with the scene's lighting and stand out in the gradient view.

Yes. No accounts, no daily caps, no per-file size limit, no watermark. The image is decoded, re-encoded, and analyzed entirely in your browser. We never see the file.

Noise and luminance analysis work on any format we can decode (JPG, PNG, WebP, HEIC). ELA only produces meaningful results on JPEGs because the test depends on the JPEG re-encoding artifact pattern.

No single tool proves manipulation. Forensic analysts combine pixel-level evidence (ELA, noise, lighting), metadata consistency (EXIF date, GPS, camera maker note), and source provenance. Use this tool to find suspicious regions, then verify against the EXIF block in our EXIF Viewer and the photo's chain of custody.

If the photo has been re-saved many times (passed through Instagram, WhatsApp, screenshots), the entire image looks bright in ELA. That is recompression noise, not manipulation. ELA is most useful on first-generation JPEGs straight from a camera or phone.

AI generators produce different noise and frequency patterns than camera sensors. The noise map and luminance gradient often look unusually clean or uniform on AI photos. Combined with a missing or generic EXIF block (no camera, no GPS, generator software in the Software tag), these are reasonable AI hints, but not proof.

The techniques are standard image-forensics methods used by FotoForensics, Forensically (29a.ch), and academic image-tampering literature. Our implementation is browser-only and open about its limits: it surfaces suspicious regions, it does not pronounce a verdict.