A RAW file is not a picture. That sounds like a technicality until the first time you try to email one, or open it on a computer that has never met your camera, and nothing works. A JPEG is a finished image. A RAW file is the sensor's notes, plus a set of instructions nobody has carried out yet.
Once that clicks, most of the confusing behaviour around RAW stops being confusing. Here is what is actually inside the file, what that buys you, and what it costs you.
What the sensor actually records
A camera sensor is a grid of light-sensitive wells. Each one counts photons and reports a single number: how much light landed here. That is all. A sensor site does not see colour.
Colour comes from a filter grid sitting on top, usually a Bayer array, where each well is covered by a red, green or blue filter. Green gets half the wells because human vision is most sensitive to green detail. So a 24-megapixel sensor does not capture 24 million full-colour pixels. It captures roughly 12 million green readings, 6 million red and 6 million blue, each one a lone brightness value with no idea what its neighbours saw.
Turning that mosaic into an image means guessing the two missing channels for every position from the surrounding wells. That process is called demosaicing, and it is interpolation. Different software guesses differently, which is exactly why the same RAW file opened in Lightroom, Capture One, darktable and the manufacturer's own converter gives you four subtly different images before you touch a single slider.
A RAW file stores the numbers from before that guess. Everything else in it is metadata: which lens, which shutter speed, what the camera thought the white balance was, and a small JPEG preview so your camera screen and your file browser have something to display.
Bit depth is the whole point
A JPEG stores 8 bits per colour channel. That is 256 possible levels of red, 256 of green, 256 of blue. It is plenty for a finished image being viewed, because that is roughly where human vision stops distinguishing steps in a smooth gradient on a normal screen.
Most cameras record RAW at 12 or 14 bits per channel, and some medium-format bodies go to 16. Fourteen bits is 16,384 levels instead of 256.
Those extra levels are not extra detail you can see. They are room to move. When you lift a shadow by two stops in an 8-bit JPEG, you are stretching maybe 20 usable levels across a large part of the picture, and you get banding and blotchy colour noise where the data ran out. Do the same to a 14-bit RAW and there were hundreds of levels sitting in that shadow to begin with.
There is a second thing going on that catches people out. RAW data is linear: the numbers are proportional to the actual photon count, not to how bright things look. Because each stop of exposure doubles the light, half of all the available levels are spent on the brightest single stop, and the darkest stops get very few. That is why "expose to the right", pushing the exposure as far up as you can without clipping highlights, produces cleaner shadows after you pull it back down. You are collecting the data in the region of the file where the levels are.
The practical result is latitude. A blown highlight that is genuinely clipped in all three channels is gone in RAW too, and no amount of dragging a slider brings it back. But a highlight that clipped in one channel only, or a face that is three stops underexposed, is very often recoverable from RAW and permanently ruined in a JPEG.
White balance becomes a free choice, not a mistake
This is the single most useful thing about RAW for most people, and it is rarely explained properly.
When your camera writes a JPEG, it decides the white balance and then multiplies the red, green and blue channels by whatever factors that decision implies. The result is baked into the 8-bit pixels. Correcting a badly wrong white balance later means stretching channels that have already been scaled and quantised, and it shows.
A RAW file does not do the multiplication. It records the raw channel values and stores the camera's white balance guess as a tag, a note that says "the camera thought this was 4300K". Your converter reads the tag as a starting point. Changing it costs nothing, because the underlying numbers are untouched. Shooting a wedding under mixed tungsten and daylight and getting the balance wrong is a fifteen-second fix in RAW and a genuine problem in JPEG.
The same logic applies to the tone curve, saturation, sharpening and noise reduction. In RAW they are instructions your software applies on the way out. In a JPEG they already happened.
Why the files are so large
The arithmetic is close to what you would guess. Every sensor site contributes one value of 12 or 14 bits, and most RAW formats compress that losslessly or almost losslessly, which typically saves a moderate amount rather than a dramatic amount. A 24-megapixel camera therefore lands somewhere in the 25 to 35 MB range per frame, and a 45 to 60 megapixel body can push past 60 or 80 MB.
A JPEG from the same camera might be 6 MB. It is smaller for two separate reasons stacked on top of each other: it has been reduced to 8 bits, and it has been through lossy compression that throws away detail your eye is bad at noticing.
Some cameras offer compressed or lossy RAW options that cut the size substantially. They are usually fine. They are also the first thing to check if you shoot RAW and are surprised by how little recovery latitude you seem to have.
Every manufacturer invented its own container
Here is where RAW stops being a format and starts being a category. There is no RAW standard. There is Canon's CR2 and CR3, Nikon's NEF, Sony's ARW, Fujifilm's RAF, Panasonic's RW2, OM System and Olympus ORF, and more.
Most of these are loosely built on TIFF-style structures, but the parts that matter, how the sensor data is packed and how the colour is meant to be interpreted, are undocumented and specific to the maker. Support in third-party software exists because people reverse-engineered it, not because a specification was published.
That has a consequence you will meet eventually: a new camera body ships and your editor cannot open its files, even though the extension is one it already supports. Manufacturers change the internal variant between models. You wait for a software update, and the update is written by someone who had to work the format out.
Now extend that forward twenty years. A file format that is only readable because of ongoing reverse-engineering effort is only as durable as somebody's willingness to keep doing that work. Open-source libraries like LibRaw and dcraw have carried an enormous amount of this load, and old formats do generally stay readable, but it is a different kind of guarantee from the one JPEG gives you. If long-term readability is on your mind, we go into it properly in which image formats will still open in twenty years.
DNG, and what it does and does not fix
DNG is Adobe's attempt at a documented, public RAW container, published in 2004 and freely available as a specification. Inside, it still holds mosaic sensor data. The difference is that the structure and the colour information are written down, so anyone can implement a reader without guessing.
A handful of manufacturers write DNG directly. Leica, Ricoh and Pentax have done so, and on phones it is common: Apple's ProRAW and Android's raw capture both produce DNG files.
You can also convert. Adobe's free DNG Converter takes a proprietary RAW and rewrites it as DNG, optionally keeping the original file embedded inside.
What DNG does not do is make the proprietary problem vanish. Converting is another interpretation step, some camera-specific processing hints do not survive the trip in the way the maker's own software would use them, and a few workflows will not accept DNG. The honest position is that DNG is a reasonable hedge rather than a solved problem, and the safest version of the hedge is to keep the original file as well rather than replacing it.
What your camera's JPEG already decided for you
When you shoot JPEG or HEIC, the camera runs a full pipeline in the fraction of a second after the shutter closes. It demosaics, applies white balance, applies a tone curve, applies its colour science, sharpens, reduces noise, converts to 8 bits, and compresses lossily.
Every one of those steps is a decision, and every one of them is effectively irreversible. Sharpening cannot be un-sharpened. Noise reduction cannot give back the detail it smoothed. Quantising to 8 bits discards the levels that would have let you push the shadows.
HEIC improves on part of this. It compresses far more efficiently than JPEG and can carry 10 bits per channel, which genuinely helps with gradients and HDR. It does not change the basic situation: the pipeline still ran, and the results are baked in.
None of that makes camera JPEGs a mistake. Modern in-camera processing is very good, and for a large amount of photography the file that comes out is better than what most people would produce from the RAW by hand. The point is only that you are accepting the camera's judgement permanently, at the moment you press the shutter.
When RAW is not worth the trouble
Shoot JPEG or HEIC without guilt when the light is controlled and consistent, when you need to hand files over immediately, when you are photographing hundreds of frames you will never edit individually, or when you are working from a phone and the storage cost is real. Family snapshots, documentation, product shots under studio lights you set yourself: the camera's decisions are fine, and RAW mostly buys you a slower workflow.
Shoot RAW when the light is difficult or mixed, when the scene has a wide brightness range, when you cannot reshoot, or when the image is going to be edited seriously. Weddings, interiors, landscapes with bright skies, concerts, anything backlit.
Shooting RAW plus JPEG is a real option and an underrated one. You get an immediately usable file and the negative, at the cost of card space.
Getting a RAW file into something you can actually share
A RAW file needs a converter, and there is a hard limitation worth stating plainly: browser-based tools, ours included, work with formats the browser itself can decode. Proprietary RAW is not one of them. Our image converter will not open a CR3 or an ARW, and no browser tool can honestly claim otherwise.
So the sequence is: develop the RAW in Lightroom, Capture One, darktable, RawTherapee, Apple Photos or your camera maker's free software, and export a TIFF, PNG or high-quality JPEG. From there the browser tools take over, and the converter is a quick way to produce a web-sized JPEG or WebP copy without a round trip through desktop software.
Keep the RAW. Export copies. The RAW is the thing you cannot recreate, and every file you make from it is disposable by design. If you are deciding what to hand to a client or upload, the format choice for that copy is a separate question, covered in JPG vs PNG vs WebP.
