Most consulate portals and biometric matching systems verify photographs against ICAO Doc 9303 and the underlying facial-image standard, ISO/IEC 19794-5. Two things tend to disqualify AI-generated portraits.

1. Face geometry that's almost-right

The standards pin down hard geometric bounds. On a 35 × 45 mm frame, the head (chin to crown) must fill 70–80% of the printed photo height — roughly 31.5–36 mm (ICAO Doc 9303 Part 3; the digital chip image in Part 9 uses a looser 50–69%). The image also needs enough resolution across the face: ISO/IEC 19794-5 requires an inter-eye distance of at least 90 pixels, and preferably 120.

Generative models produce faces with subtly inconsistent eye spacing, asymmetric ear positions, or landmarks that drift outside these bounds. Automatic validators flag it even when a human reviewer can't articulate why.

2. No AI retouching — by rule

ISO/IEC 19794-5 is explicit: the image must not be digitally altered or retouched, in particular by AI — the only technical steps it permits are cropping and background removal. A fully generated portrait fails that requirement by definition, and modern portals back it up with JPEG-level forensics (quantization tables, EXIF, and C2PA provenance markers), which carry detectable signatures even after re-saving.

What does pass

A genuine portrait, captured on a phone, against a plain background, with the crop and background correction handled in-browser by software that never alters the face itself — only crop, position, and (optionally) replace the background. That stays inside exactly the operations ISO/IEC 19794-5 allows. That's what Snapproved does.