Multiple models, in parallel
Every image is analyzed simultaneously by multiple independent AI models — three of them in thorough mode. The models come from different providers and have different strengths and weaknesses. Only when they agree do we issue a clear verdict. When they disagree, we say so honestly.
In quick mode, one model checks in about 6 seconds; thorough mode adds cross-checking and an online fact-check with sources and takes about 18 seconds.
What the models look for
- ›Hand and finger anatomy — the classic among AI mistakes.
- ›Eye reflections — light sources must be consistent in both eyes.
- ›Skin texture — too smooth is suspicious.
- ›Background transitions — blurry or illogical edges.
- ›Lighting and shadows — direction and hardness must match the scene.
- ›Text rendering — generated type often dissolves into pseudo-letters.
Every verdict lists the indicators found, so you can understand why the analysis reached its conclusion.
Online fact-check
In thorough mode, FakeCheck additionally verifies checkable claims around the image against current online sources and rates them (accurate, misleading, false, unverifiable) — including links to the sources.
What we don’t claim
Never 100% certainty. AI detection is probability-based. If confidence is below 70%, we show UNSURE — honesty matters more than a verdict that only looks certain.
A FakeCheck assessment is a machine analysis, not proof. It is no substitute for forensic examination, journalistic research, or legal advice. Generators keep getting better; what is a reliable indicator today may be outdated tomorrow. That’s why we continuously evolve our model selection.
What we don’t store
Checked images are automatically deleted from our storage after 24 hours at most. We don’t track anything per person for advertising purposes — aggregated counters for budget limits, nothing else. Details in the Privacy Policy.