AI culling
AI photo culling is in beta; learn how it flags blurry, closed-eye and dark frames, what it costs in credits and why it is not yet open in the panel.
Fonctionne en partie ou pour certains comptes ; pas encore entièrement ouvert.
La documentation est rédigée en turc et en anglais. Cette page est affichée en anglais.
Sur cette page
AI culling does the first pass before delivery for you: it measures your professional frames one by one, flags the weak ones with a reason and groups bursts. The final decision is always yours. No frame is ever deleted automatically.
This feature is in beta. The analysis and the rules are in place, but the screen where you start a run and review the results is not open in the CollectAsist panel yet. The beta sections below describe how the feature works. What you can do in the panel today is in the first section.
What you can do today
- Open and review frames one by one in the Professional tab under the event's Content tab.
- Keep a weak frame away from guests with Hide, or delete it.
- Tick several frames with Select and act on them in bulk.
What culling checksBêta
Each frame is measured separately. When measuring ends, the frames are evaluated together in shooting order.
| Reason | Explanation |
|---|---|
| Blurry | The face, or the sharpest area of the frame, is not sharp enough |
| Eyes closed | The main person's eyes are closed or half closed. Closed eyes on other people are flagged separately |
| Dark or too bright | Exposure is far too low or too high |
| Flash fault | A frame where the flash did not fire or fired late |
| Empty frame or test card | Frames with no content, and grey card shots |
| Cut-off framing | The main person's face is out of frame or cut at the edge |
| Looking away, mouth open | Passing expression problems in portraits |
| Main person's face covered | The main person's face is behind something |
| Noisy, tilted horizon | Clear noise, or a frame shot slightly tilted |
| Burst duplicate | Nearly identical frames shot within a few seconds |
Main people are worked out from the faces that appear most often at the event. The eye and expression rules are applied to these people first. The face data used for this is deleted when the analysis ends and is not kept in the culling record.
Strictness and target countBêta
You run culling at one of three strictness levels. When you change the level later, the results are recalculated; the frames are not measured again and no extra credits are charged.
| Strictness | Frames kept in a burst |
|---|---|
| Light | The best 3 frames of each burst |
| Medium | The best 2 frames of each burst |
| Strict | Only the best frame of each burst |
At the light level only clear faults are flagged. As the level gets stricter, the sharpness thresholds rise and more expression problems are taken into account.
You can also give a target number of frames. If more frames remain after the rules than your target, the lowest scoring frames are added to the rejected ones.
Reviewing the resultsBêta
When the analysis ends you get an "AI culling ready" notification in the panel. The frames are presented in three groups:
- Kept: frames with no problem found, highest scores first.
- Suggestions: frames the system has doubts about. They stay kept unless you reject them.
- Rejected: frames with a clear technical fault such as an empty frame, a flash fault or blur, and frames that have a better one in the same burst.
Review goes in this order:
- Look through the rejected frames. Move the ones you want back to kept.
- Look through the suggestions. If you like, reject every frame with a given reason in one go, for example all burst duplicates.
- Change the strictness or the target count if needed.
- Approve. After approval the rejected frames are left out of the delivery transfer and the kept ones are transferred.
Rejected frames are not deleted. You can cancel a culling run before approving it.
Over time the system adapts to your decisions. If you kept most of the frames flagged for a certain reason in your earlier runs, that reason is applied more gently for your account. Empty frames and test cards are excluded from this.
Culling is a suggestion. Important moments caught in a single frame, and frames that are blurry or dark on purpose, can be flagged by mistake. Look through the rejected frames before you approve.
Credit costBêta
| Item | Cost |
|---|---|
| Culling | 1 credit for every 100 frames started |
| Minimum charge | 10 credits per run |
Example: a run of 3,000 frames costs 30 credits, and a run of 400 frames costs 10. Credits are deducted when the analysis starts. If no frame can be analysed, the credits are refunded. Changing the strictness and keeping or rejecting single frames uses no extra credits.
For credit packages, see Credits and buying.
Review screen in the panelPrévu
A screen in the CollectAsist panel for starting a run, reviewing the results and approving them is planned.