A spot color separation webapp designed for risograph. Go to the site and start separating!
Upload an image, pick the inks you can print, and it solves for how much of each ink every color needs, then hands back one grayscale layer per ink.
How overlapping inks combine. Each is a different guess at the same physics, trading accuracy for speed.
- Subtractive — ink amounts add linearly in encoded sRGB. Not physical, but it fits nearly any target exactly and solves as a linear program, so it is by far the fastest.
- Multiply — each ink is a filter, so overlapping inks multiply and the stack darkens: yellow over blue gives green, blue over red gives a dark plum. Built from each ink's published color, so it needs no spectral data. Order-independent, and it can only darken.
- Kubelka-Munk — treats inks as filters, over 36 spectral bands from calibrated ink data, with fluorescence. Light passes down through each film, reflects off the paper, and passes back up, so overlapping inks multiply and genuinely darken. Slowest and closest to what actually prints.
- Ink Minimization — drop ink layers wherever a color can be rebuilt from fewer of them, in ΔE00. With more inks than color dimensions, many different ink combinations produce the same color; without this the solver happily spends six overlapping layers where one would do. Offered for Multiply and Kubelka-Munk: subtractive fits nearly every color exactly, so it has no metamer spread to collapse.
- Discretizations — snap opacities to a fixed number of steps, for a more posterized result.
- Ordering — which ink is laid down first. Only matters for models where overprinting is order-dependent.
- Press Simulation — screened dots spread on absorbent paper, so a nominal 50% prints closer to 75%; a dot landing on ink instead of paper spreads no further, so overprints stay where they were. Offered for Multiply and Kubelka-Munk, on by default, since without it midtones come out roughly 10 ΔE00 too light. Calibrated from one reference chart, so treat it as approximate for any particular machine and paper.
- Metallic inks are not modeled — they are flake, and reflect directionally rather than scattering diffusely. White and the grays now cover under Kubelka-Munk, but how strongly is a bracketed estimate rather than a measurement, since Riso publishes nothing about it.
- Fluorescent inks are approximated rather than modeled. Multiply gets them about right on white paper and Kubelka-Munk carries an explicit emission term, but their published colors are the least accurate in the palette — fluorescent green's is roughly 15 ΔE from what it actually prints.
- 6 inks are calibrated from real spectral measurements; the rest are still inferred from their published color. An ink Kubelka-Munk can't reproduce closely enough is left out of that model rather than rendered wrong, and selecting one drops the mode back to Multiply for you.
- Paper is assumed to be white.
- Find a way to allow zooming and panning of image.