An edited damage photo costs you a refund that was never owed. This guide shows how to vet a suspect claim photo in a few minutes — no forensic background, and without accusing anyone prematurely. The full mechanism behind it lives on our manipulated damage photo detection page.
The six on-screen visual checks
Inspect a suspect damage photo
Six checks in order. Stop the first time two of them flag.
- Find the light sourceWhere does light fall from? In a single-source scene every shadow must fall the same way with consistent softness. Two shadow directions suggest a composite.
- Compare shadow hardness to the ambient lightA scratch on a dimly lit table has a soft shadow edge. A razor-sharp shadow around the 'damage' reveals different lighting at render time.
- Check reflections on glossy surfacesPlastic, glass and varnish reflect their surroundings. If the reflection does not match the rest of the photo, that region was painted in.
- Zoom to 400% on the damage edgesEdited regions leak artefacts at the boundary: halos, repeated micro-textures (content-aware fill) or transitions that are too hard.
- Read the JPEG noise patternSensor noise is statistically uniform across one exposure. Patches with noticeably cleaner or busier noise were reprocessed.
- Align the perspectiveAn inserted defect rarely respects the surface's vanishing lines. Draw two vanishing lines — the damage should follow them.
You do not need a trained eye. Two flags out of six is enough to look closer.
The technical layer: metadata and ELA
What the eye misses often sits in the metadata: tags from an editing suite, or a completely missing camera record, are strong first-pass indicators. An error-level analysis (ELA) additionally surfaces regions compressed differently from their surroundings. How Claimscan cascades these stages without calling expensive models on every image is covered in our REST API overview.