← all demos Multimodal understanding

Multimodal Cross-Checker

Give an image and a claim about it; a vision model decides whether the image supports, contradicts, or partly supports it — and flags misleading charts (truncated axes, cherry-picked ranges) — shown as a colour-coded verdict and a confidence gauge.

How it works

  1. Input: pick a sample image (or upload one) + a claim about it.
  2. Analyze: a vision model reads the image (incl. any text/numbers) and tests the claim.
  3. Verdict: supported / partly / contradicted, with confidence + cited reasoning.
  4. Chart audit: if it's a chart, misleading techniques are flagged.
Image + claim
Vision modelread + reason
Verdict+ chart flags