An image can look ordinary while carrying faint text, words under transparency, or instructions in supported metadata.
Hidden Ink is a ufodia browser workbench for inspecting less obvious layers before an image enters an AI workflow. It combines contrast lenses, transparency inspection, simple low-bit probes, supported metadata extraction, and local English OCR.
Try it in three steps
Open the app and choose Scan the poisoned demo banner. The sample is fictional.
Switch between Original, Amplify, and Alpha dropped. Compare what becomes readable with the extracted evidence.
Open a finding, inspect its source and matched instruction rules, then export a visual view or JSON report.
Images are processed in your tab; OCR runs from bundled site assets. No account, API key, or GPU is required.
These checks can miss or misread content. They do not predict whether a particular AI will see or follow the text. Neither a clean verdict nor a transformed clean copy certifies an image as safe.
Try Hidden Ink · Explore the MIT-licensed source
If this helps you inspect an image more carefully, one coffee supports ufodia’s test cases, compatibility fixes, and documentation. Thank you.
Screenshot below: the synthetic demo banner under the Amplify lens; its instruction score is a review signal, not proof of model behavior.
