How the review order is made

Winnow creates a review order for your batch. Two distinct layers determine that order.

What gets sorted

Images with fewer recorded signals appear earlier. The order combines the label, the severity of recorded signals, and repair effort. It can cross labels — a “pass” with heavier signals may appear after a clean “needs review.” The label on the card stays the same; ordering never changes it.

Winnow provides a relative signal for setting review order. It does not determine defects or whether an image is suitable to publish. You make the final review and publishing decision.

Four review types

These four are signals recorded by the model. When a signal is recorded, its image is placed later in the review order.

Required copyMissing, typoed, or broken text
SpecsRatio, resolution, safe zones
Watermarks and fake signaturesStock marks, fake signatures
Human artifactsUnusual hands, faces, or bodies

These four signals, along with quality, logo, and policy issues, can move an image later in the review order.

Six image rules

These six are computed directly from image information. They stand apart from the judging model, cost nothing extra, and editing a rule never re-runs a judgment.

  • Unreadable image dimensions
  • Minimum short-edge size
  • Minimum long-edge size
  • Aspect-ratio range
  • Allowed file type
  • Maximum file size

What it cannot do yet

  • Your review exists only in this browser tab. Closing or reloading the tab clears its progress and results.
  • Winnow does not permanently store original images on its servers. They are sent to external processors during processing, and that retention is governed by the Privacy Policy. Judgment results and image hashes are kept for a limited period.
  • Judging is performed by an external AI provider. Cross-border transfers are stated in the privacy policy.

Sort my images

Winnow — AI image review ordering