Every AI photo tool produces the occasional below-bar image: a smoothed print, a warped hand, a garment rendered at the wrong length. The difference between tools is what happens next. Some ship it and let you burn a credit finding out. Some route it to a human retouch queue measured in business days. Quality mode does neither: it catches the defect in seconds and re-shoots, automatically.

Review one: the measurable things

Objective, instant checks run first — color similarity between your source garment and the output, a resolution floor, aspect-ratio sanity, and exposure. These are cheap and unambiguous, and they catch a class of failure no opinion is needed for.

Review two: an AI judge with your original photo in hand

Then a vision model sees your original reference next to the generated output and answers four strict questions:

Accessories get their own reviewer with presentation criteria — a handbag is carried, not worn, and judging it like clothing would fail correct shots.

The retry is steered, not a re-roll

When the review fails, the judge's named issue and the expected construction feed the second attempt — the re-shoot fixes the identified problem rather than rolling the dice again. Those retries are absorbed by CatalogX (Pro gets 3 free retries a day, Business 5), because a below-bar image should be our cost, not yours.

Fail-safe by design

If the reviewer is ever unavailable, that is never treated as a pass. And every verdict is stored with the generation, so quality is auditable over time rather than anecdotal.

Context from the market (verified September 2026): Botika's answer to output defects is a human retouch desk with a 1–5 business-day turnaround on annual-only plans. Quality mode is the self-serve version of the same promise — review and correction — at machine speed.