Combine RGB Channels and Create an LRGB Image with ImageBlend

A monochrome camera records grayscale images through different filters. The red, green, and blue masters become the three channels of an RGB color image. They need to represent the same field at the same pixel positions.

A separate luminance master, usually collected through an L filter, can supply a different or stronger record of brightness structure. An LRGB combination uses that tonal information together with the RGB color. The saved result is still a three-channel color image, not a special four-channel file.

In this lesson, ChannelCombination creates linear RGB. After the necessary processing and separate permanent stretches, ImageBlend places color from RGB over the luminance base. An LRGB blend can help when L contributes useful signal; it is not automatically better than a good RGB image.

Begin with calibrated, aligned, integrated R, G, B, and optional L masters. This guide starts after stacking. Preserve those source files and work on duplicates with clear view names such as Target_R, Target_G, Target_B, and Target_L.

All four images need matching dimensions, orientation, crop, and pixel registration. Blink corresponding stars and small structures at 1:1. Matching dimensions alone does not establish alignment; use a common registration reference and common crop if the masters differ.

The R, G, and B inputs should still be linear, without permanent stretches. They may look different under separate STFs, so do not judge their actual balance by independently auto-stretched displays.

If you have an already combined one-shot-color RGB integration, skip ChannelCombination. A separate L image can still be considered, but it must be registered to the RGB and compatible in field coverage and useful detail. This introductory exercise does not cover creating synthetic luminance or mixing Hα into RGB.

ChannelCombination assigns the input masters to channels. It does not perform registration or automatically provide physical color calibration.

Finish the appropriate linear processing on RGB, including its sky solution and SPCC calibration, before permanently stretching it. Use the linked technique lessons for prerequisites and order, especially if MGC requires SPFC first.

Process the L master for its own background, detail, and noise. Do not run an RGB color-calibration procedure on the grayscale L image just to make it match. If using MGC on L, configure its grayscale reference and flux-calibration settings appropriately; RGB settings are not a substitute.

Keep the L and RGB branches registered throughout. Do not crop, rotate, or resample only one branch. Inspect bright-star profiles and target detail before blending; a very sharp L combined with poorly matched color can expose colored fringes.

Save linear-ready checkpoints for both branches. A good RGB image is also your comparison baseline: you need it to decide whether adding L actually improves the result.

Create Target_RGB_stretched and Target_L_stretched from fresh linear-ready copies. Use the MAS lesson or another controlled stretch, inspecting faint structure and bright cores. Turn STF off afterward so you judge the actual nonlinear images.

The stretches do not have to use identical numbers. Adjust each branch so the tonal structure is useful and highlights remain manageable. Avoid forcing the histograms to look identical or making L very bright simply because it will supply the detail.

Compare the L image with the brightness structure of RGB. If L clips a core, crushes dust, or exposes harsh noise, fix that branch before attempting the color blend. ImageBlend cannot recover detail that was clipped during stretching.

For this main exercise, keep stars in both branches. If you choose a starless workflow instead, use starless L and starless RGB together and keep an aligned RGB stars branch for later restoration. Avoid accidentally mixing a starless base with RGB that still has stars.

Open Script → Utilities → ImageBlend. It is published by Cosmic Photons and includes its own documentation. Control labels may vary with the installed version; check that help when following an older screenshot.

This workflow uses luminance as the Base and RGB as the Blend color source. Colour mode is distinct from Screen mode, which is used in the separate star-restoration lesson. It is also a different method from the native LRGBCombination process; do not copy that process’s channel-weight controls into ImageBlend.

Compare the new output with RGB at the same zoom and with STF off on both. Then compare it with L to identify where the brightness structure came from. Decide whether the blend adds useful detail without making the result harsh or washing out the color.

Inspect subtle color transitions, bright-star halos, dust boundaries, and the center of the subject. Colored fringes can reveal alignment or profile mismatches; a pale appearance can reflect an overly bright luminance stretch or weak/noisy color data.

Change one part of the experiment at a time. If L is too bright, revise its stretch from the saved linear source and repeat the blend against unchanged RGB. If color needs a modest finishing adjustment, test it separately rather than increasing multiple blend and saturation controls together.

Reducing Colour blend opacity mixes more of the grayscale base into the result; it is not the same as reducing the amount of luminance added to RGB in a different blending method. Check the preview rather than treating opacity as a universal L-strength slider.

Keep the RGB version if the LRGB trial does not provide a clear benefit. More processing is not a requirement for a finished image.

If both prepared inputs were starless, the Colour blend produces a starless LRGB result. Inspect it before restoring stars; removed knots, dark holes, or poor color alignment will not be fixed just by adding stars.

Use the separately stretched, aligned RGB stars as the Blend image over the accepted starless LRGB Base in a new ImageBlend instance. Switch to Screen mode and follow the star-restoration lesson for extraction-state checks and star-strength comparisons.

Save the starless LRGB and star-restored versions separately. Keep the original extraction outputs and branch processing records. Do not add a stars layer to the main stars-intact trial, which already contains a star field.

The RGB image has bizarre colors: check the R/G/B assignments and registration before trying selective color curves. Do not use a more saturated display as proof that calibration succeeded.

RGB is almost black: check whether it is still linear and viewed with STF off. Do not permanently stretch it just to bypass a missing calibration step.

ImageBlend rejects the pair or features do not line up: confirm geometry, registration, and the intended Base/Blend selection. A blend cannot align data.

LRGB looks flat or colorless: inspect the luminance stretch, RGB color quality, Colour mode, opacity, and neutral adjustment controls. Reset a reused instance to rule out hidden filtering, masks, or a stale preview stretch.

Stars have colored rims: compare the original channel alignment and the processed L/RGB profiles. Return to a matching checkpoint instead of blurring the entire finished image to hide the mismatch.

The output has too many stars: check whether stars were present in either input before the optional restoration. Revisit the branch labels rather than simply lowering opacity.

Save the accepted output as Target_LRGB.xisf, or Target_LRGB_starless.xisf for the optional branch. Keep the RGB baseline, L input, both linear-ready files, and the ImageBlend configuration.

Record the registration reference, common crop, channel assignments, branch stretches, and blend settings. Check any inherited astrometric metadata before a later coordinate-dependent operation; creating a new image does not itself establish a valid solution.

This is an original practice sequence using the established luminance-base/Colour-blend method. The publisher’s built-in ImageBlend documentation is the reference for current controls. The linked Adam Block page lists an LRGB lesson, but its paid training content has not been reproduced here.