
Inside every RAW file from a modern camera is a full size JPEG the camera rendered at the moment of capture, the same one you see on the back of the camera. Most culling tools ignore it and decode the RAW from scratch for every frame. That single choice is most of the difference between a cull that finishes inside your card copy and one that takes the evening. This article explains the trade and where it stops being a good one.
A RAW is sensor data plus metadata plus, in nearly every format that matters, a preview JPEG at or near full resolution: CR3, NEF, ARW, RAF, DNG and ORF all carry one. The camera made it so that the rear screen and the in camera playback could show you something instantly. It reflects the picture style, white balance and exposure the camera used, which is exactly the view you had when you decided to keep shooting.
Reading that JPEG is cheap. Decoding the RAW means demosaicing the sensor data, applying the colour pipeline, and producing an image you can look at. On the same machine it is up to forty times the work for a first look that, for culling purposes, gives the same answer.
| Reason a frame is set aside | Settled from the JPEG? | Why |
|---|---|---|
| Eyes closed | Yes | A face and its eyes are the same at preview size as at full RAW resolution. |
| Same shot (burst duplicates) | Yes | Similarity between frames does not need sensor data. |
| Nobody in frame | Yes | Presence is a coarse question. |
| Very dark or blown | Mostly | The JPEG shows what the camera's render clipped. A RAW may hold more, which is why borderline frames go to the real file. |
| Out of focus | Mostly | Sharpness is measurable on the preview for all but the closest calls. |
On the reference wedding of 4,812 frames, Kepla opened the real RAW on 214 of them: the borderline exposure and focus calls, plus every frame it flagged as uncertain. The other 4,598 were settled from the render the camera had already made.
The JPEG is a render with the camera's choices baked in. Two things follow. First, a frame that looks two stops under in the preview may be perfectly recoverable in the RAW, so a culler that trusts the JPEG for exposure verdicts will clear frames you could have saved. That is why the borderline ones go to the RAW. Second, a blur that is on purpose looks the same as a blur that is a mistake in either file. No amount of decoding settles taste, which is why Kepla marks those frames uncertain and asks you.
The honest design is not JPEG or RAW. It is JPEG first, RAW where the answer needs it, and a question where the answer is yours.
Cost per frame is only half the story. Bytes per frame is the other half. During a card copy the drives are the bottleneck, and a tool that reads the whole RAW for its own purposes competes with the copy for the disk. The embedded JPEG is a fraction of the file, so it fits in the idle gaps while the copy runs. That is how the sort on the reference wedding finished 19 minutes 48 seconds before the copy to two drives did.
Ask how many frames of a typical wedding the tool decodes in full, and whether it runs at full speed while an import is in progress. A tool that decodes everything will answer the first with all of them and the second with a pause. Kepla answers 214 of 4,812, and yes, inside the copy. It is free through the private preview and the first hundred photographers keep it at $99 a year. Request Founding Access.
For blinks, duplicates and empty frames, no: the answer is the same at preview size. For borderline exposure and focus, Kepla opens the RAW. On the reference wedding that was 214 frames of 4,812.
All of the ones Kepla reads: CR3, NEF, ARW, RAF, DNG, ORF, plus JPEG and HEIF captures themselves.
No. Reading the embedded preview is a read only operation. Decisions are written to XMP sidecars beside the file.
Kepla for Mac clears the obvious misses from a card, names the reason on every frame it sets aside, and leaves the choosing to you. Nothing is ever deleted, moved or renamed. Free through the private preview · the first hundred photographers keep it at $99 a year.