Color Space Profiles
Built-in Color Profiles
Siril contains a basic set of color space profiles. These are taken from Elle L. Stone's excellent repository of free and well-behaved color profiles. Her website contains a wealth of technical information on color space theory and practice, and a great source of inspiration for applying color processing to images. All the profiles there are released under the Creative Commons Attribution-Share-Alike Unported license, version 3.0.
The built-in profiles are intended to provide a sufficient set of color spaces for most purposes: sRGB for web export and as a default display profile, Rec2020 as a wide gamut profile that looks like it may become the next standard for high quality wide gamut monitors, and of course a range of Gray profiles to match the RGB ones.
sRGB
Linear profile sRGB_elle_V4_g10
sRGB TRC profile sRGB_elle_V4_srgbtrc
ICC sRGB Preference profile with perceptual lookup tables (for display only)
sRGB TRC profile sRGB_elle_V2_srgbtrc (for export)
Rec2020
Linear profile Rec2020_elle_V4_g10
Rec709 TRC profile Rec2020_elle_V4_rec709
Rec709 TRC profile Rec2020_elle_V2_rec709 (for export)
Gray
Linear profile Gray_elle_V4_g10
sRGB TRC profile Gray_elle_V4_srgbtrc
sRGB TRC profile Gray_elle_V2_srgbtrc (for export)
Rec709 TRC profile Gray_elle_V4_rec709
Rec709 TRC profile Gray_elle_V2_rec709 (for export)
Note that the native TRC versions of the profiles are provided in two versions (version 4 and version 2). Version 4 provides better functionality (especially for high bit depth images) including parametric TRCs which avoid quantization. However version 4 profiles are not universally supported yet. So when exporting a file for use in other software it is safest to embed a V2 profile.
The Gray profile is available with TRCs to match the two built-in RGB color spaces, and also with a linear TRC.
Linear TRC Profiles
Sensor data starts off as a linear representation of light, so why don't we assign a linear color profile to unstretched images such as a newly stacked stack? Well, technically perhaps we should. Certainly if you wish to do so you can, and nothing bad will happen. You can do this using the color management dialog, and there is also a preference (Pedantically Assign Linear ICC Profiles) that will assign a linearized version of the working ICC profile when a profile is auto-assigned to a newly loaded image (if it shows no sign of having been previously stretched), a stacked image or a newly composited image. However there is generally no benefit in doing so. To understand why, we need to understand a fundamental difference between astrophotography and normal photography processing.
In normal photography the whole image is well exposed, with darker shadow regions and lighter highlights. At the raw stage the image is converted from linear light to a working profile, but this is reversible and in fact is frequently reversed as many editing operations (color blending, noise reduction, etc.) are best done in linear light.
However in astrophotography, all the detail is very deep in the shadows and we have to apply a very strong stretch indeed in order to produce an image that looks pleasing. This stretch is much more extreme than the mild change of gamma from 1.0 to 2.2 going from linear light to sRGB, and it is not easily reversible - especially if multiple stretches are applied. So we do still need to do the same kind of operations (color correction, noise reduction, deconvolution etc.) on linear light (and a lot of other things that don't even exist in normal photography should be done to linear images too: star modelling, star removal etc.)
It isn't necessary to set a linear color profile to do these operations - algorithms like noise reduction etc. aren't even aware of color profiles. They just apply themselves to the provided data.
So it is safe to assign the color profile to an image at any stage of editing, though it usually makes most sense to do it just before stretching. The key thing to remember is that with astrophotography, all the operations that need to be done on linear images should be done before any stretching. Generally, stretching and any final tweaks to color balance, saturation etc. should be just about the last editing operation that is done to an image. And at that point you should be editing it in your chosen color space, which will give as consistent as possible a look to the image when viewed in any color managed application or output device.
Color Space Recommendations
This section explains in a bit more detail why Siril offers the built-in color spaces it does.
sRGB is the de facto standard for the web. As mentioned above, if you want to guarantee your images look good in all web browsers, as well as all third party applications that may or may not support color management, you really need to export in sRGB. It's a good fit for most SDR monitors, and in the absence of proper color management images that looked right in previous versions of Siril should be considered to be in sRGB.
However, consider the CIE 1931 color space again. The horseshoe shape forming the boundary of the color space represents pure spectral tones. If you have a perfectly monochromatic light source like a laser or one of those sodium yellow street lights, you're looking at the boundary of CIE 1931. As astrophotographers these pure monochromatic sources are actually really important to us, especially if you do narrowband imaging. Those colors can't be represented accurately in sRGB. The proportion of visible colors that can be represented in sRGB is actually rather small. We can do better.
So, if a wide gamut is better why is the wide gamut color space built into Siril Rec2020 rather than an even wider one? AllColorsRGB, ACES 2065 and ProPhoto RGB are all much bigger - ACES 2065 can represent all parts of the visible spectrum.
The problem is, the way they do that is by setting their primary colors - the "100% red", "100% green" and "100% blue" values - outside the visible range. This is problematic especially for narrowband imagers who want to assign filters to primary colors: you end up with imaginary colors in your image and always have to rely on the intent of a color space transform to do the right thing in turning them into something visible. Also some of these color spaces are linear, such as ACES 2065. This is good for CGI artists but not so good for us, because of the slow display transforms when working with linear spaces. Siril can optimise some transforms from linear color spaces over and above what lcms2 does, but only if the RGB primaries are the same, e.g. if transforming linear sRGB to g22 sRGB). So Rec2020 was chosen as it provides a nonlinear color space with the widest gamut without having imaginary primaries.
My recommendation, if you aren't already sold on a particular color space, is to try Rec2020 as a color space for editing, as well as for sending images to high-quality printing services that can cope with color managed images, and sRGB for web export. But there are also other good choices available using ICC profile files that you can use instead, such as Adobe RGB and ProPhoto RGB (ROMM), if you prefer.
Third Party Color Profiles
Monitor Profiles Siril includes the ICC's v4 sRGB monitor profile with support for Perceptual rendering LUTs and uses this by default. A different monitor profile may be used in its place, but note that only the intents supported by a given profile will be available - many widely available sRGB profiles only support the Relative Colorimetric intent.
Soft Proofing Profiles There are a wide variety of press standards and papers in the world and each combination requires its own ICC profile. Siril cannot provide all of these. Therefore in order to use the soft proofing display mode you will need to provide the appropriate soft proofing ICC profile in the Preferences window.
You may wish to work in a color space other than the inbuilt ones, for example ProPhoto RGB. This is supported, but you will need to provide ICC profiles yourself. The Preferences window provides controls to set a RGB profile and a Gray profile with a matching TRC (essentially, with the same gamma). If you intend to export files in your chosen color space it is recommended that the profile you provide is a V2 ICC profile, for maximum compatibility with other software.
The image shows how to set up a ProPhoto working space using Elle Stone's profiles. The standard gamma of ProPhoto RGB (ROMM) is 1.8, so as well as the ProPhotoRGB profile, we add the Elle Stone Gray profile with gamma = 1.8. You can download all of Elle Stone's profiles here. If you don't have the ProPhotoRGB profile, you can also use Elle Stone's "LargeRGB-elle-v2-g18" profile, which is exactly the same except she has avoided use of the term "ProPhoto RGB" for possible copyright reasons.
HDR Display Limitations
Wide gamut display color spaces such as Rec2100 with HLG or PQ TRCs may need greater than 8 bit pixel buffers to display smoothly. Unfortunately, Siril uses the Cairo graphics library for display and Cairo cannot yet process pixel buffers wider than 8 bits. The impact of this is likely negligible for most users. However, if you are lucky enough to be using a >1000nit HDR display capable of displaying wide gamut color spaces and are using certain combinations of third party image and custom monitor profiles then for certain images, you may experience minor color staircasing artefacts. These are only a display issue and won't affect the printed appearance of your image, or even its appearance when transformed to a narrower gamut or viewed on operating systems that do support high bit depth pixelbuffers.
For the moment there isn't anything we can do, however if Cairo adds high bit depth support in the future then there may be room for improvement.