AI photo culling, explained

“AI culling” gets used to mean everything from a blur filter to a full editing pipeline. Here’s what it actually is, how it works, and what separates a genuinely useful tool from a black box.

What culling is

Culling is the step between import and editing: choosing which frames are worth keeping, ranking the keepers, and discarding the misses. For a high-volume shooter — weddings, events, real estate, sports — it’s often the most time-consuming part of the job, and the least creative.

AI culling automates the mechanical parts: detecting technical problems (soft focus, motion blur, blown exposure, closed eyes), grouping near-duplicates, and ranking frames so the strongest rise to the top.

How quality scoring works

A good culling model doesn’t return a single mysterious “good/bad” verdict. It scores each frame on several independent dimensions and combines them. LightVision uses seven:

• Clarity — overall sharpness and micro-contrast
• Focus — is the subject actually in focus
• Open eyes — are the subjects’ eyes open
• Warmth of emotion — expression and connection
• Composition — framing and balance
• Energetic moment — is something happening
• Exposure — is the frame well-exposed

These combine into a 0–100 score and a Pick / Alternate / Review recommendation. The reason multiple dimensions matter: a frame can be tack-sharp and perfectly exposed but emotionally flat, or slightly soft but capture the moment. A single score hides that trade-off; named dimensions let you see it and override it.

Explainable beats opaque

The most important question to ask of any culling AI: can you see why it scored a frame the way it did?

If the tool just hands you a ranking, you’re trusting a black box — and you’ll second-guess it, which defeats the time savings. If it shows per-dimension scores, you can trust the Picks at a glance and know exactly where to look when you disagree (e.g. “great moment, low focus — I’ll keep it anyway”).

Cloud vs on-device

Culling tools run in one of two places:

• Cloud: your images (or derivatives) are uploaded to a server for processing. Convenient, but it means your clients’ photos leave your machine, and you depend on a connection and a vendor’s uptime.
• On-device: the models run locally on your computer. Nothing has to leave your machine, there’s no per-image upload wait, and it works offline.

For client work — especially weddings and anything with privacy expectations — on-device is the safer default. LightVision runs culling, grading, style learning and semantic search on-device by default; a cloud model is optional, never required.

What to look for

When evaluating an AI culling tool, check:

1. Explainability — per-dimension scores you can inspect, not a single opaque rank.
2. Duplicate & burst grouping — moments collapsed into one decision.
3. Person awareness — different people never merged; the same person tracked across the shoot.
4. Where it runs — on-device vs cloud, and whether cloud is optional.
5. What happens next — does the cull flow into editing and delivery, or dead-end into another tool?

The goal of AI culling isn’t to replace your eye. It’s to do the mechanical 90% — the grouping, the technical checks, the ranking — so your judgment goes where it actually matters.