RGB to CMYK: what really happens to your colours in print

Why there are two colour spaces at all

A screen makes colour by emitting light. Every pixel is three tiny lamps, red, green and blue, and each lamp can be off, fully on or anywhere in between. Turn all three up and you get white; turn them all off and you get black. That is RGB: additive colour, built from light.

Paper does the opposite. It starts white and gets darker as ink is laid on it. Cyan, magenta and yellow inks each absorb part of the light that hits the paper, and black ink (the K, for key) is added because the three coloured inks together make a muddy dark brown rather than a true black, and because black ink is cheaper. That is CMYK: subtractive colour, built by taking light away.

The two systems are not two names for the same thing. The full set of colours a device can show or print is called its gamut, and the gamut of a press on paper is much smaller than the gamut of a screen. Backlit pixels can be pushed to a purity that reflected ink can never reach, because paper only sends back the light that falls on it and the inks are not perfectly pure pigments. However good the press and however glossy the stock, a real share of the colours you can see on a monitor simply do not exist in ink.

This is why converting to CMYK is not a file-format change like saving a PNG as a JPG. It is a translation into a language with fewer words. Every colour outside the ink gamut has to be replaced by the nearest colour that is inside it, and the software doing the replacement has to decide what nearest means: keep the hue and lose saturation, keep the lightness and shift the hue, or compress the whole image slightly so the relationships between colours survive. None of those choices is free, and none of them can be undone once the RGB original is gone.

It also explains something that confuses a lot of people the first time they order print: the file looked fine on screen, the printer did nothing wrong, and the poster still came back duller. Nobody made a mistake. The image contained colours that paper cannot hold, and nobody checked before the file went out.

Which colours fall off first

The losses are not spread evenly. Most of an ordinary photograph, skin, sky, foliage, wood, stone and fabric, sits comfortably inside the CMYK gamut and prints very close to how it looks on screen. The trouble is concentrated in a few families of colour, and they are the ones designers reach for when they want something to pop.

• Electric and royal blues. Pure RGB blue is the single worst case. Cyan and magenta ink mixed together lean towards purple, and the bright, deep blue of a screen has no ink equivalent at all. Blue logos, blue gradients and blue neon signage routinely come back violet or slate.
• Fluorescent and lime greens. A screen green at full saturation is far outside anything cyan and yellow can mix. Bright green text, eco branding and sports-kit greens flatten into a duller olive or grass green.
• Very saturated oranges and reds. Magenta plus yellow gives a good red and a decent orange, but the glowing, sunset-warm orange of a phone screen loses a visible step of intensity. Safety-vest orange and heavily saturated sunsets are the usual victims.
• Deep purples and magentas. Less common in photographs, very common in branding. They tend to lose depth and go slightly greyer.

Neutral tones are affected in a different way: not lost, but shifted. A clean grey built from equal RGB values becomes a mix of four inks on paper, and if the profile is wrong or the press drifts, that grey picks up a cast, usually warm or slightly green. That is why greys and skin tones are the first thing an experienced printer looks at on a proof.

You do not have to guess which of your colours are at risk. Almost every serious image editor has a soft-proof mode, sometimes called proof colours or print preview, that simulates a chosen CMYK profile on screen without changing the file, and most pair it with a gamut warning that paints every out-of-gamut pixel in a flat grey or a loud colour so you can see exactly where the problem is. Turn both on with the profile your printer uses (more on that below) and look at the image at 100 %. If the warning lights up a logo, a sky or a background you care about, you have three honest options: desaturate that area a little yourself while you can still see the original, pick a slightly different colour that sits inside the gamut, or accept the shift. What you should not do is find out from the finished print.

The ICC profile: what it is and why you keep it

An ICC profile is a small block of data that describes what the numbers in an image actually mean. The values 255, 0, 0 are not a colour; they are an instruction, and the colour they produce depends on which red the device in question can make. The profile is the dictionary that turns the numbers into a definite colour in a device-independent reference space, and from there into the right numbers for the next device in the chain.

Most images on the web and from phone cameras are tagged sRGB, the narrow, standard space that ordinary screens are built around. Cameras set to a wider space, and many images from professional workflows, are tagged Adobe RGB (1998) or ProPhoto RGB, which can hold colours that sRGB cannot. Print-ready files carry a CMYK profile that describes a particular combination of press, ink and paper: ISO Coated v2 (FOGRA39) and PSO Coated v3 (FOGRA51) are the common European standards for coated paper, uncoated stock has its own, and North American printers often work to GRACoL or SWOP. Other regions have their own reference conditions too.

The profile matters at every hand-over. When an image travels from a phone to a laptop, from an editor to a layout program, from the layout program to a PDF and from the PDF to the printer's RIP, each program reads the embedded profile to interpret the numbers correctly. If the profile has been stripped somewhere along the way, the next program has to assume something, and its assumption is usually sRGB. For an sRGB file that is harmless. For a wide-gamut file it quietly desaturates the picture. For a CMYK file opened by something that expected RGB, it can make everything look wrong at once. The colours in the file did not change; the meaning of the numbers did.

The same logic applies to converting. Converting to CMYK means picking a destination profile, and the right destination is whichever profile describes the press and paper your job will actually use. Convert to a generic CMYK from a menu and you are guessing at the printer's ink limits, dot gain and paper white. Guess wrong and the printer either has to convert again, from a file that has already lost colours it can never recover, or prints your guess as it is. So before you convert anything, ask two questions: which profile do you want, and do you want the file in CMYK at all? Many commercial printers prefer to receive a clean, profiled RGB file and convert it themselves on a calibrated workflow, because that keeps every option open until the last moment. If they do want CMYK, they will name the exact profile and usually offer it as a download. Use that one and nothing else.

A last practical point: keeping the profile costs nothing. It adds a few kilobytes to the file. Stripping it saves nothing worth having and creates exactly the kind of colour shift this whole article is about. Whatever tool touches your image between camera and press, check that it writes the profile back out.

What AI does here, and what it does not

There is a temptation to bundle everything under one heading: the image is being prepared for print, so an AI upscaler must be involved in the colour side too. It is not, and it should not be.

Upscaling adds pixels. Denoising removes grain. Sharpening recovers edges. Restoring an old photo rebuilds faces and fills in damage. Every one of those operations works on the image's detail, and every one of them can and should happen in RGB, in the image's original colour space, with the original profile still attached. None of them needs to know what paper you will print on, and none of them changes the meaning of the colour numbers. Run them in RGB and you keep the widest possible gamut all the way through, which is exactly what you want: the more colour information the file still has when the final conversion happens, the better that conversion can be.

The colour conversion is a different job with a different owner. It is done either by your editor, with the printer's named profile and a rendering intent you have chosen, or, more often, by the printer on their own calibrated system. It belongs at the end of the chain, after every other edit, because it is the one step that throws information away.

This is how Airpic is built. The Image Upscaler, and the Photo Denoiser, Photo Unblur and Old Photo Restorer alongside it, all work in RGB and write the output back with the ICC profile the input carried. A file tagged Adobe RGB goes in tagged Adobe RGB and comes out tagged Adobe RGB, with more pixels. EXIF orientation is applied so the result is the right way up, and the tools accept JPG, PNG (including 16-bit and alpha), WebP, TIFF, HEIC and AVIF, so a wide-gamut TIFF from a scanner or a HEIC straight from a phone can go in as it is. Airpic does not convert to CMYK, and does not claim to: that step needs the printer's profile, and only the printer knows which one it is. What the AI gives you is a file with the resolution the print needs and the colour information intact, ready for whoever performs the conversion last. Your file is deleted an hour after you download the result, a day at most, and is never used for training.

Step by step: resolution first, profile intact, CMYK last

Here is the order that avoids the surprises, whether you are printing a 10 × 15 cm photo or a 3-metre banner.

1. Ask the printer three things before you touch the file. The resolution they want at the final size (300 ppi is the usual figure for anything viewed at arm's length; large-format work seen from a distance needs far less), whether they want RGB or CMYK, and, if CMYK, the exact profile. Write the answers down.
2. Work out how many pixels you need. Multiply the final width in centimetres by the required ppi and divide by 2.54. A 30 × 45 cm print at 300 ppi needs roughly 3,540 × 5,310 pixels. Compare that with the pixel size of your file. The large-format printing guide has a calculator that does this for you, including the viewing-distance adjustment for posters and banners.
3. Fix the image before you enlarge it. Noise, blur and damage get bigger along with everything else. Run denoise, unblur or restoration first, on the original, in RGB.
4. Check the gamut while you still have the original. Soft-proof with the printer's profile, turn on the gamut warning and decide what to do about anything that lights up. Adjusting a saturated blue by hand now is better than letting a blind conversion decide for you.
5. Upscale to the target size. Open the Image Upscaler, drop the file in and choose the factor that gets you to the pixel count from step 2: ×2 or ×4 for most jobs, and the large-format print mode when you need to go further, up to 600 megapixels processed as tiles. Switch on face enhancement if there are people in the shot and the faces are small; leave it off for landscapes and products. For flat illustration or line art, use the Anime Upscaler instead, which is built for clean edges rather than photographic texture. Processing and the preview are free; when the result looks right, download the full file. The profile is still in it.
6. Save without losing anything. Deliver a TIFF or a maximum-quality JPG at the final pixel size, with the embedded profile. Do not run it through a web compressor: that step is for screens, not paper.
7. Convert to CMYK last, or not at all. If the printer asked for RGB, you are done: send the file. If they asked for CMYK, convert in your editor using their named profile, check the soft proof once more and export. Keep the RGB master. If a second run on a different paper comes along next year, you will convert from the master again, not from a file that has already been through one conversion.

The whole sequence takes a few minutes. The only part that is genuinely hard, deciding what to do with a colour that paper cannot print, is the part you can now see coming. When the file is clean and the profile is in place, start with the Image Upscaler.

FAQ

Keep reading