Convert RAF to PNG

When a photograph becomes a product, it passes through more software than the person who took it usually realises. The platform makes a preview, the layout engine places it into a template, a proof is generated for you to approve, and the production file is rendered for the press. If your upload was a JPEG, several of those steps are re-encodes, and each one starts from the artefacts the previous one introduced.

None of that is visible on a screen and some of it is visible on paper, particularly in smooth gradients and around high-contrast edges — a horizon, a bright window, lettering on a sign. Uploading a lossless file does not stop the platform re-encoding; it stops the chain being longer than it needs to be, and it is the only part of the process the photographer has any say in.

The converter finds the largest JPEG Fujifilm embedded in the raw file and writes it into a PNG. That preview is frequently smaller than the full frame, because the camera wrote it for its own screen rather than as a deliverable, so convert one and read the dimensions before assuming you have every pixel the sensor recorded. That picture carries the film simulation, the white balance, the tone curve, the grain effect and the noise reduction the camera applied. It is not a development of the sensor data.

For this particular use that is unusually defensible. A photographer choosing Fujifilm for Classic Chrome or Acros is choosing a rendering, and the camera’s version of that rendering is the reference one. What you cannot do is change it afterwards: the simulation is baked in, an Acros frame stays monochrome, and the highlight and shadow latitude that makes raw worth shooting is not in the extracted picture at all.

A print-on-demand product has a physical size and a printing resolution, and between them they decide how many pixels are useful. A 6 by 4 inch postcard at 300 ppi needs 1,800 by 1,200. An A4 print needs about 3,500 by 2,480. A 12 by 12 inch calendar page needs 3,600 square. An X-T5 frame is 7,728 pixels across, which is more than any of those and considerably more than most.

Set a maximum width during the conversion to the number the product needs, with a little margin for the manufacturer’s bleed. It reduces the file by the square of the reduction — cutting 7,728 to 3,600 removes about four fifths of the pixels — which is what turns an upload that fails a size cap into one that succeeds, at no visible cost on the finished object.

PNG compresses by finding repetition, and a photograph has almost none. A 26-megapixel preview is 78 MB of raw pixels and typically comes out between 35 and 60 MB; a 40-megapixel one is 119 MB raw and lands roughly between 55 and 90. Read those as the full-resolution case, since the preview is often smaller than the frame. The range within it is wide because it depends on the picture: a studio shot on a plain background compresses well, a woodland scene barely compresses at all.

The encoder here runs an oxipng optimisation pass after the initial encode, which costs about half a second per file and removes roughly two thirds of the bytes the encoder alone would produce. That is a substantial saving on a file of this size and it is applied automatically — there is no setting to enable, and the pixels are identical either way.

Every pixel value stored is recovered exactly, and a PNG resaved a hundred times is bit-for-bit the file it started as. That is the property being bought here and it is a real one. Each chunk also carries a CRC, so a file corrupted in transfer fails loudly rather than displaying with a band of garbage down one side — useful when a large upload goes over a hotel connection.

What it does not mean is that nothing was lost. The picture inside the RAF was compressed by the camera when it wrote the preview, so the PNG is a perfect copy of something that had already been through one lossy stage. Lossless describes the container’s relationship to its contents, not the contents’ relationship to the sensor.

PNG supports 16 bits per channel, which sounds like a good home for a 14-bit Fujifilm capture. It is not, because the 14-bit data never reaches the encoder: the embedded preview is an eight-bit JPEG, so the PNG written from it is eight bits with no more tonal range than a web image.

For a print product that is generally sufficient, and it stops being sufficient if the picture will be graded before it goes to press. A 16-bit file that genuinely carries the sensor’s range has to come from a raw developer working on the RAF — Lightroom, Capture One, darktable or Fujifilm’s own software will export exactly that, and it is a different job from the one this page does.

The alpha channel is PNG’s best-known feature and it is empty here. Every pixel recorded by a sensor is fully opaque, so there is nothing transparent in the file for the PNG to carry, and the channel exists in the container without holding anything.

If the product needs a cut-out — an object floating on white, a subject over a coloured background in the template — that is a masking job in an image editor after this conversion, and the PNG is the right format to save the result in. The conversion gives you a container that can hold transparency; it does not create any.

No ICC profile is written, so anything opening the file will assume sRGB. Fujifilm bodies have a colour space menu offering sRGB and Adobe RGB, and if the camera is set to Adobe RGB the preview holds Adobe RGB numbers that will be read as sRGB ones — which shows up as saturated colours coming back tamer than intended.

On a make chosen for its colour rendering that is a particularly frustrating way to lose the look, and it is easy to miss until a proof arrives. Print-on-demand platforms almost universally expect sRGB anyway, so setting the camera to sRGB for work destined for print products is the simplest fix, and checking the menu costs nothing.

The conversion decodes and re-encodes, so no metadata reaches the output: no camera or lens, no exposure settings, no capture date, no copyright field and no GPS coordinate. PNG can hold EXIF and a copyright text chunk; this conversion writes neither.

The copyright one is worth planning around if you are selling the work. Attribution has to travel in the platform’s own listing fields, in the file name, or in a visible credit, because nothing inside the file claims authorship. The absence of a location tag is the compensating gain: a photograph taken in your own garden no longer ships its coordinates with the product listing.

The RAF is read from disk by the page, scanned for its embedded picture, decoded and written back out as a PNG — all in the tab, with no request carrying the image. The network tab shows it and a disconnected machine proves it.

That matters when the pictures are a product that has not gone on sale. Preparing a print run through a free converter that uploads means the images exist on somebody else’s server before they exist in your shop, under terms you did not read. Here they do not leave the laptop, and a batch of forty frames is limited by your own processor rather than by a queue.