Simulate real-world social media compression, filtering, and re-encoding attacks to measure SynthID retention.
Simulate real-world social media compression, filtering, and re-encoding attacks to measure SynthID retention.
This inspector uses procedural images and preset metadata to demonstrate watermark and provenance concepts. The spectral heatmap is simulated. Its confidence score is calculated from authored formulas, and its signature labels come from preset flags. It does not verify a real uploaded image or cryptographic signature. For a synthetic injected preset, set blur, crop, and noise to zero. Reducing JPEG quality from one hundred to fifty gives degradation equal to fifty times zero point four five, or twenty two point five. The confidence formula subtracts zero point seven two times degradation, moving from ninety four point two to seventy eight percent. The page calls a signal detected at fifty percent confidence or above. Its provenance panel separately reads validity and tamper flags from the selected preset. Compare the channels and export the report, but do not treat these simulated labels as proof of real authorship or authenticity.