Humans identify AI content correctly 28% of the time. Worse than a coin flip. ScanTrace does it in 15 seconds — forensic analysis across three independent layers, with a written explanation of every verdict. We return 3 verdicts: REAL, AI_GENERATED, or INDETERMINATE — because when evidence is weak, the honest answer is we don't know. Built for journalists, fact-checkers, and photo editors. Free: 10 analyses/month, forever. Pro: 400/month · video analysis · PDF certificates.

Sydekiq — Your own AI agent, deployed on a private server, 24/7

Your own AI agent, deployed on a private server, 24/7

Hey PH 👋 I'm Santiago. I built ScanTrace alone over 2 months — and I want to tell you about the decision that almost didn't happen. Every advisor said the same thing before launch: don't return INDETERMINATE. Binary results convert better. Users want certainty. They were probably right about conversion. But when forensic evidence is genuinely too weak to decide, returning FAKE with 52% confidence means the first time someone checks manually and we're wrong, the tool gets abandoned. Trust is harder to rebuild than to lose. So we kept INDETERMINATE. ~15% of real-world images land there. I think that's the most honest thing we built. Three independent forensic layers, a written explanation with every verdict, and 97–99% accuracy across the major models. Free plan: 10 analyses/month. Try it on an image you're not sure about. If you've ever published — or almost published — something that turned out to be AI-generated, I'd genuinely love to hear about it. That's exactly why this exists. 👇