What Is a Keep-Reject-Review Workflow?

Beyond a simple binary sort

The simplest way to sort a shoot is binary: every frame is either a keeper or a reject. That works fine for the frames that are obviously one or the other, a black frame, a tack-sharp portrait, but it forces a firm answer on the frames that are not obvious at all. A keep-reject-review workflow adds a third bucket for exactly those cases, so a genuinely uncertain call is not quietly forced into keep or reject just because the system only offers two options.

Why the third bucket helps

Forcing a low-confidence decision either way hides the fact that it was low-confidence in the first place. A borderline frame sorted straight to reject might be a real photo lost to an overcautious system; sorted straight to keep, it might let a genuine miss slip into the gallery. A dedicated review bucket keeps those borderline calls visible, so a human makes the final decision on exactly the frames where a measurement alone should not be trusted.

A concrete example

Picture a formal portrait where one guest's eyes look almost, but not quite, fully open. A binary sorter has to guess: keep it and risk a flawed group shot, or reject it and risk losing the only usable frame of that grouping. A keep-reject-review workflow instead routes that one frame to review, sorts the clearly closed-eyed frames to reject and the clearly open-eyed frames to keep, and leaves you with a short, specific list of genuine judgment calls instead of a guess buried in either pile.

Limitations

A three-way sort is only as good as the confidence measure deciding what counts as uncertain. Set the bar too loose and almost nothing reaches review, which defeats the purpose; set it too tight and review fills up with easy calls that did not need a second look. A review bucket also still needs someone to actually work through it, since an ignored review folder provides no real benefit over a binary sort.

How Cull AI Studio handles this

This three-way sort is Cull AI Studio's core mechanic. Every frame in a scan is sorted into KEEP, REJECT, or REVIEW, both inside the app and as folders on disk, Save, REJECT, and Review. REVIEW is gated by a confidence score, with an uncertainty bar of 0.65 by default: anything below that bar is routed to REVIEW instead of being silently sorted into KEEP or REJECT. Originals are never edited or deleted regardless of which folder they land in, and every decision is recorded in the project's audit trail so it can be checked or reversed later.