Can I Adjust the AI's Confidence Threshold?

What the confidence bar actually does

Every decision Cull AI Studio makes, sharpness, exposure, faces, eyes, combines into a single confidence value for that frame. When the app is confident a frame is a clear keep or a clear reject, it sorts it accordingly. When the confidence falls below the uncertainty bar, 0.65 by default in the app's scoring, the frame goes to REVIEW instead of being pushed into KEEP or REJECT on a guess. That threshold is a property of the app's scoring, not a dial in the interface you turn up or down before a cull.

Why that is worth being honest about

It would be easy to promise a more aggressive or more cautious mode and call it a threshold slider. Cull AI Studio does not currently expose the underlying number as a user setting, so the honest answer is that you are not adjusting a threshold directly. What you are doing instead is deciding, wedding by wedding, how much of the REVIEW pile you clear versus how much you trust, and feeding the software corrections that shift what it considers confident over time.

A concrete example

Say you shoot a dim reception where the dance floor lighting confuses exposure and sharpness scoring across a whole batch of frames. Rather than the software confidently keeping or rejecting all of them, a large share lands in REVIEW because their confidence sits under the bar. You are not fighting a setting to loosen up the cull for this batch; you are working the queue the software already built for exactly this situation, deciding each frame with the context the AI does not have.

How you actually influence how cautious the AI is

• Work the REVIEW queue deliberately. Every confirm or reversal is recorded, and reversals carry double weight when your personal model trains.
• Give reasons when you correct a call. Reasons such as closed eyes, missed focus, or better expression become part of what the model learns from.
• Let the personal model train. Once roughly 1,000 reviewed decisions are on file, a model trains on your computer and starts reflecting your own tolerance for borderline frames.

Honest limitations

Because the threshold itself is not a setting you can move, you cannot simply tell the software to be more conservative on a given wedding and have every borderline frame flagged instead of decided. The review pile reflects the app's own uncertainty, not a preference you set in advance. If you want the AI's confident decisions themselves to shift toward your standards over time, that comes from accumulated corrections and training, not from a configuration change made before a cull.

How Cull AI Studio handles this

Cull AI Studio's scoring routes anything below its default 0.65 confidence bar to a REVIEW folder and queue instead of deciding it silently, so uncertainty is visible rather than hidden inside a confident-looking KEEP or REJECT pile. Every decision and every correction, with the reason you give, is recorded in the project's audit trail, and reversals carry double weight when your personal model trains locally on your computer after about 1,000 reviewed decisions. A new model is tested against your own held-out past decisions and never silently replaces the working one, so as your corrections accumulate, the software's confident calls move toward your standards even though the underlying number is not something you turn by hand. For more on reading and trusting those scores, see what a culling confidence score is.