Convert NEF to JPG

Every Nikon writes a complete JPEG into the NEF at the moment of capture. It is what the rear screen shows, what a file browser uses for a thumbnail, and on modern bodies it is the full pixel dimensions of the sensor. Extracting it needs no decoder at all — the picture is already a picture, it simply has a raw file wrapped around it.

That is what this page does. It scans the NEF for embedded JPEGs, validates each one, and returns the largest by pixel count. What it does not do is demosaic: reconstructing colour from the Bayer grid, applying a tone curve and a colour profile is a different and much larger piece of work, and NX Studio, Lightroom and darktable are where it belongs.

Nikon has been writing NEF since 1999, longer than any other format in this group, and a working archive frequently spans a D70, a D750 and a Z6. Those files are all called NEF and they are not the same internally. Sensor resolution, bit depth, compression scheme and — the part that matters here — the size of the embedded preview all changed across that span.

Recent bodies embed a full-resolution JPEG, so the extracted picture is the same pixel dimensions as the raw. Older ones embedded something smaller. Nothing warns you which is which until you convert, so on a mixed archive it is worth checking the dimensions of a few outputs before assuming the whole set came out at full size.

Cameras usually embed more than one JPEG: a small one for the file browser, sometimes a mid-size one, and the large view. The extractor keeps the largest above 640 by 480 and ignores anything smaller, because handing somebody a 160-by-120 thumbnail and calling it a conversion is worse than telling them it did not work.

On the oldest Nikon files that floor is occasionally the whole story — the file has a preview and it is too small to be useful, so the conversion reports a failure. That is the correct outcome and the right next step is a raw decoder, which reads the sensor data and does not care what the preview looks like.

Nikon has offered several ways to store the sensor readings: uncompressed, lossless compressed, a lossy compressed mode often described as visually lossless, and on the Z8 and Z9 the High Efficiency and High Efficiency star options that cut file sizes dramatically. Which one was used has a real effect on file size and on how much latitude survives.

None of it changes this conversion, because the sensor data is not read. The embedded preview sits alongside it as an ordinary JPEG regardless of how the raw was packed, so a High Efficiency NEF from a Z9 and an uncompressed NEF from a D850 extract in exactly the same way.

The workflow this suits is the one most photographers already run informally: make everything viewable, decide what matters, then do careful work on the few frames that earned it. Extracted JPGs are perfect proofs — they are the camera’s own rendering, which is what the client saw on the back of the camera and what they are expecting.

The NEF files stay where they are. Converting produces a second file and consumes nothing, so there is no migration decision to make and nothing to regret later. The rule of thumb is to treat the JPGs as disposable and the NEFs as the negatives.

The Picture Control is baked in. If a frame was shot in Monochrome, the JPG is monochrome and the colour cannot be brought back from it — the colour is in the NEF. Auto white balance that went wrong under mixed lighting is likewise fixed in place, and correcting it in an 8-bit JPEG costs more than correcting it against 14-bit sensor data.

Highlight recovery is the clearest case. A sky that reads as blank white in the extracted preview usually still has detail in the NEF, because the preview was rendered with a tone curve that clipped it. That detail is one reason to shoot raw and it is not something this conversion can hand you.

Because the preview is already a JPEG it could be handed over byte for byte, metadata and all, which makes this the one pair on the site where keeping the shooting data is even possible. It is not what happens by default. "Remove metadata" is on, the EXIF block is edited out, and the JPGs you send carry no lens, no exposure figures and no coordinates.

That default is there because location data travels further than anyone intends. A Nikon with built-in GPS, or one paired with a phone or a logger, writes coordinates into the file, and a proof set reaches more people than a delivered edit does. But for proofing the exposure information is genuinely useful — it is right there when a client asks why one frame looks different, and the timestamps keep the set in order — so turn the option off for a client set and leave it alone for anything going public. Setting a maximum width is not the way to strip metadata any more: it only ever did so by forcing a re-encode, which spends a generation of quality on something the option does directly. The ICC colour profile stays in both cases.

Extraction is cheap. There is no decoder to run and, for a JPG target at the default settings, no encoder either — the preview is located and copied out. That is why a set that would take a raw developer an hour goes through here in a fraction of it, and why there is no daily allowance to work around — the batch takes a hundred files at a time, and a four-hundred-frame set is four passes rather than a rationed one.

The free tier accepts files up to 100 MB each. An uncompressed D850 NEF is around 90 MB, which is the closest anything in this bundle comes to the ceiling; lossless compressed files from the same camera are roughly half that.

The conversion happens in the browser. For a photographer this is not an abstract benefit — an unpublished shoot, a wedding not yet delivered, a commercial set under embargo — none of it is uploaded, because there is nowhere for it to be uploaded to.

It also means the tool works on a laptop in a hotel with poor wifi, on a machine where you cannot install software, and at an airport where you would rather not push ninety gigabytes over a shared connection. The engine downloads once; the photographs never move.