Camera RAW Explained: What’s Actually Inside a CR2, NEF, or ARW

RAW files are not images — they are sensor measurements waiting to be interpreted. What that means for editing latitude, colour, and converting to JPEG.

A RAW file is not a picture

This is the idea everything else depends on, and it is genuinely counterintuitive. When you open a JPEG, you are looking at an image: a grid of pixels each carrying a red, green, and blue value, ready to display. When you open a RAW file, there is no such grid. What exists is a set of single-number brightness readings, one per photosite on the sensor, and each of those photosites sat under a coloured filter.

Camera sensors cannot see colour. They count photons. To recover colour information, manufacturers cover the sensor in a mosaic of red, green, and blue filters — usually a Bayer pattern, with twice as many green filters as red or blue, because human vision is most sensitive to green. Every photosite therefore records the brightness of exactly one colour, and knows nothing about the other two.

Turning that mosaic into a normal image is called demosaicing, and it is fundamentally an act of interpretation: for every pixel, two of the three colour values have to be reconstructed from the neighbours. Different software makes different decisions, using different algorithms. This is why the same RAW file opened in Lightroom, Capture One, and your camera maker’s own software produces three visibly different images, all of them legitimate.

What the extra bits actually buy you

A JPEG stores 8 bits per colour channel, giving 256 brightness levels. A RAW file stores 12 or 14 bits — 4,096 or 16,384 levels. That difference sounds academic until you try to rescue a photograph.

Consider a shot where the sky is blown out. In a JPEG, the highlight information was compressed into a handful of values at capture time, and pulling the exposure down produces flat grey where clouds should be, because there is nothing left in the file to recover. The same recovery from a RAW often brings back full cloud structure, because those levels were recorded and simply were not being displayed in the preview you were looking at.

White balance works the same way. A RAW file stores the white balance as a metadata note rather than applying it to the pixels, so changing it afterwards is completely free and lossless — exactly as if you had set it correctly in camera. In a JPEG the white balance was baked in, and correcting a heavy colour cast means stretching channels that have already been quantised into 256 steps, which introduces banding and colour noise.

The same logic covers shadow lifting, sharpening, and noise reduction. RAW keeps the decisions open; JPEG closes them at the moment of capture.

Why every manufacturer has its own format

Canon writes CR2 on older EOS bodies and CR3 on everything since 2018. Nikon writes NEF. Sony writes ARW. Fujifilm writes RAF, Olympus ORF, Panasonic RW2, and there are dozens more. Worse, the internal structure changes between camera generations, so software has to be updated for each new body released.

None of these formats are publicly documented. Support in third-party software exists because developers reverse-engineered it, camera model by camera model, which is why a photo tool can open RAW files from one body perfectly and refuse files from a newer one released last year. Updating your software usually resolves it, but the underlying fragility is real.

Adobe published DNG in 2004 as an openly documented alternative anyone can implement, and archivists took it seriously — it is a recommended preservation format at several national institutions. Adoption among camera makers was thinner than hoped: Leica, Ricoh, Pentax and the drone manufacturers adopted it, while Canon and Nikon never did. Where DNG genuinely won was mobile, which is why an iPhone shooting ProRAW or an Android phone in RAW mode hands you a .dng file.

The practical consequence is that RAW is a poor archival format on its own. If long-term readability matters, keep the original RAW and store either a DNG copy or a high-quality export representing how you actually processed the image.

Why your converted RAW looks flat

This is the most common surprise, and it is not a fault. Your camera screen showed you a vivid, contrasty, sharpened image — but that was a JPEG preview embedded inside the RAW file, generated by the camera with the manufacturer’s colour science, tone curve, and your selected picture profile applied. The RAW data underneath is deliberately neutral.

When conversion software renders the actual sensor data rather than extracting that preview, the result looks flatter and less saturated by comparison. That flat starting point is the entire point of shooting RAW: it is the room you have to make your own decisions, rather than a finished picture someone else already decided for you.

Convertmaxxing renders RAW with a neutral profile rather than attempting to imitate any particular vendor’s colour science, and says so plainly. For a quick shareable copy that is usually fine. For an image you have carefully graded, export from your RAW editor so its interpretation is preserved — converting the original RAW will not reproduce the edits you made elsewhere.

A practical workflow

Shoot RAW when the situation is difficult or the image matters: tricky lighting, high contrast, uncertain white balance, or anything you plan to edit seriously. Shoot JPEG when volume and immediacy matter more than latitude, and the lighting is straightforward. Many cameras will record both at once, which is a reasonable hedge if you have the card space.

Do your exposure and colour adjustments in a RAW editor, while the 12 or 14 bits of latitude still exist. Then export. Converting a flat RAW straight to JPEG and trying to fix the exposure afterwards means working with 256 levels per channel instead of 16,384, and the recovery that would have been effortless becomes banding and noise. Convert as the last step, never the first.