How the Dad or Mad AI Likeness Scan Works

No math degree required. Here's what happens between choosing your photos and opening the family-likeness report.

"The AI compared your faces" can sound like magic. In practice, Dad or Mad uses a short preparation stage and a private AI review to turn a photo set into a structured, just-for-fun family-likeness report. Here's the current flow without the jargon.

1. Check whether each photo is usable

First, browser-based checks answer practical questions: is there one clear face, does it fill enough of the frame, and is the photo reasonably sharp and front-facing? If the app finds no usable face, several faces, heavy blur, or a difficult angle, it asks you to replace that photo before processing.

2. Prepare a consistent photo set

The app keeps the photo you selected first as the primary frame, includes up to a few usable photos per person, and resizes them for a reliable upload. That selected-first ordering stays consistent across the preview, saved report, and visual crop boards so the evidence does not quietly switch photos between sections.

3. Review the visible resemblance privately

When you run the preview, the selected photos are sent for private AI processing. The model reviews the set together and describes visible similarities and differences across features such as the eyes, brows, nose, mouth, face shape, and other observable traits. Dad or Mad does not use that review to establish identity, ancestry, or a biological relationship.

4. Turn observations into a structured report

The AI returns a defined report structure rather than a loose paragraph: overall family-likeness strength, scan reliability, parent lean when both parents are present, feature-by-feature attribution, and analyst notes. The app validates that response before showing or saving it, so missing or malformed sections are rejected instead of being filled with made-up browser values.

5. Keep strength separate from reliability

A likeness score describes the visible resemblance; scan reliability describes how steady that signal is from the supplied photos. Clear, front-facing, well-lit images and agreement across multiple frames improve reliability. Blur, strong expressions, age gaps, shadows, and side angles can lower it. Neither number is a probability of parentage.

That's it — readiness checks, photo-set preparation, private AI review, schema validation, and an honest reliability rating. See the privacy policy for retention and deletion details. And remember: this comments on how faces *look*, not genetics — see can a face tell you who a child looks like? for the important limits.