Image Blend

Purpose and expected result

Image Blend combines a base image A with a same-size blend image B using one of 27 pixel or color-component modes. Each input can receive a linked midtones transfer, the blend image can receive an optional filter, and the mode result can be attenuated by opacity and the base image's active mask. OK destructively replaces the base target after a successful run; the observable checkpoint is a new Image Blend history step and a nonlinear target.

At a glance

Item	Current behavior
Menu path	Process > Utilities > Image Blend
Enabled when	At least one image view is active; the process does not require exclusive window access.
Inputs	Base still/sequence plus one other open image of exactly equal width and height. A grey base and a colour blend image, or the reverse, blend as the engine's composition promotes them; the second image is read in its own units whatever its representation, and the result keeps the base's representation.
Outputs	Destructively replaces the base still or sequence after success. Take snapshot creates an auxiliary still. No file is written directly.
Recommended image state	Depends on intent. Arithmetic combination of calibrated linear data can be meaningful, but the process always marks a successful result nonlinear. Contrast/color modes, input MTFs, and color filters are normally nonlinear operations.
Preview support	Checked by default; asynchronous full-image preview with histogram, original/result switch, display autostretch, and snapshots. Preview area is supported: the second image is cropped to the same area.
Sequence support	Applies each base-sequence frame against the second image taken from its view when Apply is clicked; every frame is blended as a still image of it would be. There is no paired-frame synchronization.
Mask support	Yes. The active base mask is applied after mode and opacity: ordinary mask white selects the processed result, black keeps the original; inversion reverses this.
External dependencies	None. Gaussian Blur and High Pass are the engine's exact Gaussian, at any radius; Hue and Color are the engine's colourizing filters.

Before you start

Register and crop both inputs to the same pixel coordinates. The blend selector checks dimensions only; it does not compare WCS, distortion, channels, exposure, background, or photometric scale. Duplicate the base or save a checkpoint because OK is destructive.

Preview area crops the blend image to the same area as the base, so an area preview shows the blend of that area.

For linear combination, start with identity stretches (Black Point 0, Highlights 1, Midtones 0.5) and Filter Type: None. Align and normalize externally when required; Image Blend performs neither operation.

Procedure

1. Activate the image to be replaced and open Process > Utilities > Image Blend. Confirm it in Base image > Image.
2. Select an equal-size registered view in Blend image > Image. Its blend, filter, and opacity controls become enabled.
3. Leave both Stretch groups at identity initially. Choose Replace to verify source selection, then select the required Blend mode.
4. Keep Filter Type: None until the unfiltered equation is understood. Add Gaussian Blur, High Pass, Hue, or Color only to B, before blending.
5. Reduce Opacity from 1 to mix the mode result back toward original A. If the base has an active mask, remember that the mask attenuates the result once more after opacity.
6. Use Preserve lightness only when it is enabled and you want the mode result reconstructed with base CIE L*. Inspect gamut clipping and color at 100%.
7. Compare PREVIEW with BASE IMAGE using Switch image under one display stretch. Inspect the result histogram; display Autostretch does not change output samples.
8. Select OK. Wait for processing to finish; the base is replaced and its history records the mode and selected blend image.

Complete parameter and action reference

UI label/action	Type and unit	Runtime default	Persisted-default exception	Valid values, step, or constraints	Effect of increasing, decreasing, or selecting	Dependencies and enable/visibility rules	Quality, artifact, runtime, memory, or disk trade-off	Scope: preview/apply, channel, mask, still/sequence	Persistence
Histogram plot	Graph, normalized intensity/count	Current preview channels	None	Bins set by Plot resolution	Shows processed, opacity- and mask-aware preview distribution	Present while process is open	Recalculation adds work	Preview display only	Recomputed
Histogram left-drag	Pointer gesture	Idle	None	Press/drag	Shows bin coordinate/count readout	Histogram shown	Diagnostic only	One plot	Not persisted
Histogram middle-drag	Pointer gesture	Idle	None	Press/drag	Pans a zoomed plot	Histogram shown	Diagnostic only	One plot	Window lifetime
Zoom in	Histogram action	Fitted	None	1.30x per click	Magnifies plot	Histogram toolbar	More scrolling	Display only	Window lifetime
Zoom out	Histogram action	Fitted	None	0.70x per click	Shrinks plot	Histogram toolbar	Less detail	Display only	Window lifetime
Zoom reset	Histogram action	—	None	One click	Fits plot	Histogram toolbar	Display only	Display only	Immediate
Show grid	Toggle	Checked	None	Off/on	Shows plot grid	Histogram toolbar	Display only	Display only	Window lifetime; Reset unchanged
Show clipping info	Toggle	Checked	None	Off/on	Shows endpoint counts/percentages at normalized <=0 and >=1	Histogram toolbar	Detects clipping in preview samples	Display only	Window lifetime
Absolute scales	Toggle	Unchecked	None	Off/on	Checked normalizes every visible channel to its own peak; unchecked uses a shared peak	Histogram toolbar	Checked conceals relative peak-count strength	Display only	Window lifetime
Draw histogram areas	Toggle	Unchecked	None	Off/on	Filled additive areas versus lines	Histogram toolbar	Filled overlap can obscure bins	Display only	Window lifetime
Show histogram	Toggle	Checked	None	Off/on	Shows/hides plot and disables/enables related controls	Histogram toolbar	Can reduce redraw distraction	Display only	Window lifetime
Plot resolution	Bin-count list	8-bit (256)	None	64, 128, 256, 512, 1024, 4096, 16K, 64K, 100, 1000, or 10000 bins	More bins retain narrower distribution detail	Histogram shown	Higher computation/memory; output precision unchanged	Display only	Window lifetime
Base image > Image	Target selector	Active image	None	Any open target accepted by launch guards	Supplies A, mask, metadata base, history, and destructive destination	Changing target clears Blend image selector	Wrong target is replaced	Preview/OK; still/sequence	Target lifecycle; Reset unchanged
Base image > Stretch > Black Point	Double + slider, normalized	0.0	None	0..1, four decimals, spin step 0.01	Raises lower clipping boundary of linked MTF	Always enabled	Clips more low samples	All base channels before mode; mask later	Captured; Reset/success 0
Base image > Stretch > Highlights	Double + slider, normalized	1.0	None	0..1, four decimals, step 0.01	Lowers upper clipping boundary when reduced	Always enabled	Clips/compresses high samples	All base channels before mode	Captured; Reset/success 1
Base image > Stretch > Midtones	Double + slider, MTF midpoint	0.5 identity	None	0..1, four decimals, step 0.01	Below 0.5 brightens interior samples; above darkens	Always enabled	Extremes collapse most interior values	All base channels before mode	Captured; Reset/success 0.5
Blend image > Image	Equal-size view selector	No image	None	Other open view matching base width/height	Supplies B; enables remaining blend controls	Rebuilt on view lifecycle/target changes; no channel/type/WCS/intensity validation	Misregistration creates doubled edges/seams	One fixed image for preview/OK and every base-sequence frame	Window lifetime; target change clears
Blend image > Stretch > Black Point	Double + slider, normalized	0.0	None	0..1, four decimals, step 0.01	Raises B MTF lower boundary	Valid B required	Clips B shadows	All B channels before filter/mode	Captured; Reset/success 0
Blend image > Stretch > Highlights	Double + slider, normalized	1.0	None	0..1, four decimals, step 0.01	Reducing lowers B upper boundary	Valid B required	Clips/compresses B highlights	All B channels before filter/mode	Captured; Reset/success 1
Blend image > Stretch > Midtones	Double + slider, MTF midpoint	0.5 identity	None	0..1, four decimals, step 0.01	Below 0.5 brightens B; above darkens	Valid B required	Strong nonlinear redistribution	All B channels before filter/mode	Captured; Reset/success 0.5
Filter Type	Combo box	None	None	None, Gaussian Blur, High Pass, Hue, Color	Chooses preprocessing applied only to stretched B	Valid B required; reveals Radius or Hue/Saturation as appropriate	Adds buffers and processing; changes subsequent equation input	Before blend/opacity/mask	Captured; Reset/success None
Radius	Double + slider, Gaussian sigma / high-pass radius	10.0	None	0.1..1000, three decimals, step 0.01	Larger Gaussian removes larger detail; larger High Pass isolates coarser residuals	Visible only for Gaussian Blur/High Pass	Large values cost more and can create broad halos/boundary effects	B only; exact Gaussian, then mode	Captured; Reset/success 10
Saturation	Double + slider, normalized colorizer saturation	1.0	None	0..1, three decimals, step 0.01	Raises strength of chosen fixed hue	Visible only for Hue/Color filters	High values can clip gamut	B-filter output before mode	Captured; Reset/success 1
Hue	Double + slider, degrees	0.0	None	0..360, three decimals, step 0.01	Rotates fixed filter color; 0/360 are red	Visible only for Hue/Color filters	Strong color replacement; output clipped	B-filter output before mode	Captured; Reset/success 0
Opacity	Double + slider, blend fraction o	1.0	None	0..1, four decimals, step 0.05	0 returns original A; 1 uses full mode result; interior linearly interpolates	Valid B required	Lower values soften all mode/filter/MTF effects	After mode/Preserve lightness; before active mask	Captured; Reset/success 1
Blend mode	Combo box, 27 choices	Replace	None	Every mode in the equations table below	Selects function F(A,B)	Valid B required	Contrast modes clip readily; Lab/HSI modes cost more	Before opacity/mask	Captured; Reset/success Replace
Preserve lightness	Checkbox	Unchecked	None	Off/on	After mode, replaces result CIE L* with stretched base L*	Disabled for Color, Lightness, or mono+mono; slider tracking pauses while checked	A colour conversion each way is costly, which is why slider tracking pauses; gamut can clip; opacity/mask can alter final measured L* afterward	RGB result before opacity/mask	Captured; Reset/success off
Preview	Checkbox	Checked	None	Off/on	Creates/destroys live private preview updates	Available while idle	Checked recomputes on changes	Preview only	Window lifetime; Reset delegates to shared preview reset
Preview pane	Viewer	Current result	None	Pan/scroll plus the toolbar controls listed below	Displays processed preview	Preview checked	Holds result/statistics buffers	Display only	Window lifetime
Fit To Window	Preview-toolbar action	Preview fitted initially	None	Click or Ctrl+Shift+F	Fits the whole preview image	Preview checked	Broad context; resampling can hide pixel-scale artifacts	Display only	Preview lifetime
Zoom In	Preview-toolbar action	Current scale	None	Click; 1.25× relative zoom	Enlarges around the viewport center	Preview checked	Better local blend/halo inspection	Display only	Preview lifetime
Zoom Out	Preview-toolbar action	Current scale	None	Click; 0.8× relative zoom	Reduces the displayed scale	Preview checked	More context, less pixel detail	Display only	Preview lifetime
Zoom 100%	Preview-toolbar action	Current scale	None	Click; absolute scale 1.0	Shows one-pixel viewer scale	Preview checked	Preferred for clipping, halos, and registration residuals	Display only	Preview lifetime
Switch image	Preview action	PREVIEW shown	None	Click/Space/right-click/Ctrl+Shift+O	Toggles PREVIEW and BASE IMAGE	Preview checked	Display only	Preview	Window lifetime
Preview area	Preview checkbox	Unchecked	None	Off/on; move/resize overlay	Crops A and the selected B to the same area and shows their blend there	Preview checked	A smaller area is less work	Preview/snapshot only, never OK	Window lifetime; Reset may not clear
Take snapshot	Preview action	No snapshot	None	Click/Ctrl+Shift+K	Re-executes current settings and opens a new still	Preview active	Full-size allocation/processing; active mask applies	Auxiliary still; target unchanged	One view per action
View Channel	Preview-toolbar combo box	Follows the current result channel layout	None	RGB, grayscale, stored channels, and supported derived CIE/HSL/HSV/HSI/YCbCr display components	Changes only the displayed component	Preview active; choices follow result channels	Helps inspect per-channel or derived-component artifacts	Display only; output/histogram samples are unchanged	Preview lifetime
Display Mode	Preview-toolbar combo box	Linear	None	Linear or Autostretch	Linear shows stored blend samples directly; Autostretch derives a display presentation from result statistics	Preview active	Autostretch can conceal absolute changes or clipping	Display only; output/histogram samples are unchanged	Preview lifetime
Autostretch Settings	Preview-toolbar action	Initialized from the target view	None	Opens shadow clipping, target background, saturation, brightness, and contrast settings	Changes only Autostretch presentation/statistics	Preview active; relevant when Display Mode is Autostretch	Adds display computation; blend output is unchanged	Display only	Preview lifetime
OK	Dialog action	Available even before B selection	None	One click	With valid B, replaces target after success; without B, applies Base Stretch alone to A - one screen transfer on each colour channel - records the result nonlinear under image-blend, and describes the step with its three values, such as Base stretch: shadows 0.000, midtones 0.250, highlights 1.000; at the identity values (Black Point 0, Midtones 0.5, Highlights 1) A is left as it was	Target lock required; only target/shared preview actions are generically disabled during work	Full-image buffers; destructive; most Image Blend controls remain live though run parameters are captured	Still/sequence, mask-aware	Success creates history then resets process values
Reset	Dialog action	Available	None	One click	Restores mode, stretches, filter, opacity, and preserve	Does not change target/B selectors or histogram state	Schedules preview	Configuration	Immediate
Close	Dialog action	Available	None	One click	Cancels the previews and, for a sequence run, the progress dialog, which stops that run; while a run is still in flight the window hides and closes once the run has landed, so a still OK already running still lands	Remains live during processing	Nothing waits on the GUI thread, so closing never blocks the window	Process/window	No settings persist after destruction
Sequence progress / Cancel	Shared progress	Created for sequence OK	None	Per-frame progress; cooperative cancel between/within supported stages	Writes temporary replacement, commits only on success	Base sequence only; B is fixed	Full sequence I/O and buffers	Sequence, opacity/mask per frame	Cancel removes temporary output

How the algorithm works

Conceptual pipeline

1. Read base A and selected blend image B. Apply each image's linked MTF when its controls differ from identity.
2. Apply the selected filter to B only. A grayscale image meets an RGB one as three equal channels, except a grayscale B in Lightness, whose samples are the lightness itself. Both images are read as fractions of full scale - an unsigned integer as a fraction of its largest value, a signed integer counted from its smallest, a floating-point image as it is - and an integer base is worked in a Float32 copy (Float64 for 32-bit integers).
3. Evaluate the selected mode R = F(A,B). When requested and supported, reconstruct RGB R using base CIE L* and result a/b.
4. Apply opacity: Q = (1-o)A_original + oR. If an active ordinary mask M exists, output O = (1-M)A_original + MQ; an inverted mask exchanges those weights. Write every sample once in the base's representation, rounded to nearest and saturating; build the result's metadata from both images, mark the output nonlinear, and commit only after success.

Input stretch and filters

For normalized x, black point s, highlights h, and midtones m, linked MTF is

T(x)=0 for x<=s, T(x)=1 for x>=h, otherwise with u=(x-s)/(h-s), T(x)=((m-1)u)/((2m-1)u-m).

All channels use the same parameters. The UI does not enforce s<h; crossed endpoints produce a degenerate clipped mapping rather than a useful stretch. Identity (0,1,0.5) skips the allocation.

• Gaussian Blur: the engine's exact Gaussian with standard deviation Radius, at any radius; one above about 32 pixels runs as several exact passes whose variances add up to it.
• High Pass: clamp(B - Gaussian(B,r) + 0.5, 0, 1).
• Hue: if B is RGB, first reduce it to normalized CIE L*; then use that scalar as HSV Value with selected Hue/Saturation. Mono follows the same scalar colorization.
• Color: apply the same fixed-HSV mapping directly to B. RGB is processed channel-wise, so it is not a conventional single-hue recoloring unless B is mono.

Blend-mode equations

The table uses per-channel values A,B read as fractions of full scale, as described above; clamp means [0,1]. Every mode - Replace, Darken/Min, Lighten/Max, Linear Dodge/Add, and Subtract included - is computed on those fractions for every representation, signed integers included, and written in the base's representation. RGB Darker/Lighter Color select the entire pixel by CIE L*; component modes are defined separately.

Mode	Current implementation F(A,B)	Interpretation / boundary
Replace	B	Replaces A before Preserve lightness/opacity/mask; the result is written in the base's representation, not B's
Darken/Min	min(A,B)	Per-channel darker sample
Multiply	A*B	Darkens; white is neutral, black returns black
Color Burn	0 if B=0, else 1-min((1-A)/B,1)	Strong shadow/contrast burn
Linear Burn	clamp(A+B-1)	Additive darkening with black clipping
Darker Color	Mono min(A,B); RGB copies the complete pixel with lower CIE L*	Avoids channel-by-channel mixed color for RGB
Lighten/Max	max(A,B)	Per-channel lighter sample
Screen	1-(1-A)(1-B)	Lightens; black is neutral
Color Dodge	Current code: 0 if B=1, else clamp(A/(1-B))	The exact-white B special case returns black, unlike the limiting/conventional white result
Linear Dodge/Add	min(A+B,1), computed in double for every representation	Brightens with white clipping
Lighter Color	Mono max(A,B); RGB copies the complete pixel with higher CIE L*	Whole-pixel L* selection
Overlay	2AB if A<=0.5, else 1-2(1-A)(1-B)	Threshold is base A
Soft Light	clamp((1-2B)A^2+2AB)	Quadratic contrast response
Hard Light	2AB if B<=0.5, else 1-2(1-A)(1-B)	Threshold is blend B
Vivid Light	0 if B=0; max(0,1-(1-A)/(2B)) if B<0.5; otherwise min(1,A/(2(1-B)))	At exact B=1, division tends to infinity and clamps to 1 for positive A
Linear Light	clamp(A+2B-1)	Current B=0 special case is also 0
Pin Light	clamp(max(2B-1,min(A,2B)))	Current B=0 special case is 0
Difference	abs(A-B)	Identical registered samples become 0
Exclusion	clamp(0.5-2(A-0.5)(B-0.5))	Lower-contrast difference family
Subtract	max(A-B,0), computed in double for every representation	Clips negative results to black
Divide	0 if B=0, else clamp(A/B)	Useful for ratios/flatness only after valid normalization
Power	clamp(pow(A,B))	B<1 generally lifts nonzero A; implementation follows standard pow boundaries
Arctan	clamp(atan2(A,B)) in radians	Values above 1 radian clip; it is not normalized by pi/2
Hue	HSI (H_B,S_A,I_A) converted to RGB	Grayscale inputs are promoted to neutral RGB
Saturation	HSI (H_A,S_B,I_A) converted to RGB	Takes saturation only from B
Color	CIE Lab (L*_A,a*_B,b*_B) converted to RGB	Preview and OK use the same conversion
Lightness	CIE Lab (L*_B,a*_A,b*_A) converted to RGB; mono B maps B*100 to L*	Replaces luminosity while retaining base chroma; Preview and OK use the same conversion

Preserve lightness performs a second Lab reconstruction after F: it uses (L*_A,a*_R,b*_R). Color already does this and Lightness intentionally replaces L, so the checkbox is disabled for those modes. Opacity and mask blending occur later in RGB/sample space, so final measured CIE L is not guaranteed to remain exactly L*_A when either is fractional.

Boundary and failure behavior

• Only width/height are filtered in the selector. The run uses the image the blend view showed when the run was asked for; one whose size no longer matches the base is refused with the blend image does not have the target's size.
• Each image must have one or three channels. A grayscale base and an RGB blend image, or the reverse, give an RGB result; a grayscale B in Lightness is read as the lightness itself.
• The result is always in the base's representation, whatever the blend image's; an integer result is rounded to the nearest code and saturates.
• Arithmetic/color functions clamp where the table says. Linear Dodge/Add and Subtract hold the sum at one and the difference at zero, in double, for every representation. A pixel either image's validity rejects keeps the base's samples; a sample that is not a finite number in a pixel its image states valid makes the engine refuse the run.
• Preview and OK run the same composition, so Darker/Lighter Color, Color, Lightness, Preserve lightness, and the Hue filter's L* use the same CIE conversion in both, computed in double through the engine's XYZ matrices.
• A run the engine refuses publishes nothing; the shared framework reports why and keeps the original target.

Parameter interactions and starting points

Combination	Conservative starting point	Too weak looks like	Too strong looks like	Adjustment order
Difference for registration	Identity MTF, None filter, opacity 1, no mask	Residuals hidden by low opacity	Broad residuals from scale/background mismatch dominate	Register/normalize first, inspect Difference, then lower opacity only for context
Screen/Add for emission/stars	Identity MTF, opacity below 1	Added structure is barely visible	White clipping, enlarged stars, bright background	Match geometry/scale, choose mode, lower opacity, then mask
High Pass + contrast mode	Radius 10, moderate opacity	Little local detail transfer	Halos/noise and clipped contrast	Set radius, mode, opacity, then mask
Preserve lightness + color mode	Use Hue/Saturation modes; opacity 1 initially	Color change subtle	Gamut clipping or false color	Choose component mode first; use Preserve only where enabled; then opacity
Active mask + opacity	Tune unmasked at low opacity, then enable mask	Effect absent in selected area	Hard mask transitions or excessive local effect	Tune equation/filter, opacity, then mask softness/inversion

Inputs, outputs, and metadata

On a successful still run, the base view is replaced and receives a history step named Image Blend with <mode> with <blend-view-name>. Output geometry and representation are the base's. The result has three channels when either image has three or the mode or filter makes colour, and one otherwise. CFA mosaics are treated as grayscale sample planes, and the result states no color filter array.

The result's metadata is built from both images. Its exposure is the sum of the exposures the two images state - even for Difference, Replace, low opacity, or a mask - recorded once, and the images' EXPOSURE and EXPTIME cards are dropped. The object, instrument, telescope, observer, filter, sensor temperature, gain, binning, and exposure start are kept when both images state the same value and dropped otherwise; every other header card of the base is kept only when the blend image states it identically. The base's world coordinate (WCS) cards and plate solution are kept, because the pixel grid is the base's, without checking registration. The output processing state is then set to nonlinear with image-blend provenance, and the lineage names the engine composition.

For a base sequence, one fixed blend view image is reused for every frame and a temporary sequence replaces the original only after success. It does not pair two sequences by index or time.

Performance

Core per-pixel modes are parallel and scale with width, height, channels, and sequence frames. Linked MTF and filters add full-image buffers; Gaussian/High Pass add neighborhood work; CIE/HSI modes and Preserve lightness add conversions. Each preview and run applies the transfers again; nothing is cached between runs. Plot resolution affects histogram cost, not blend precision. No production timing or peak-memory measurement was made.

Base Stretch on a sequence

Every frame of a base sequence is stretched by the Base Stretch values itself.

Worked example

• Evidence mode: Source-and-test-derived.
• Input condition/provenance and settings: same-size registered pair; identity stretches; Filter None; Difference; Opacity 1; Preview on; no active mask; Preview area off.
• Expected checkpoint: aligned equal pixels map to black under abs(A-B); unequal structure appears as a positive residual; reducing opacity mixes that residual back toward original A.
• Inspect at 100%/statistics/history: inspect stars/edges at 100%, clipping and background in the histogram, then verify the Image Blend history description.
• Limitations: not executed on approved astronomical data; no alignment accuracy, visual quality, sequence output, timing, or memory claim.

Warnings and limitations

• OK is destructive. Save or duplicate the base first.
• Equal dimensions do not prove registration or matching WCS. There is no alignment, resampling, background match, or normalization.
• Preview area crops both images to the same pixel area; it does not register them.
• B is one fixed image - the one its view showed when the run was asked for - not a synchronized sequence.
• Color Dodge returns black at exact B=1 in the current implementation.
• Most Image Blend-specific controls remain enabled during execution. The run uses the parameters and the blend image captured when it was asked for.
• The result keeps the base's representation whatever the blend image's; integer results are rounded to the nearest code. A sample that is not a finite number in a pixel its image states valid makes the run fail.
• Metadata exposure summation does not account for blend mode, opacity, or mask; the base's WCS is not validated against B.

Troubleshooting

Symptom or exact error	Verified cause/validation path	Recovery	Evidence source
Blend image is absent	It is the base itself or has different width/height	Register/crop/resize first, then reopen selector	isViewControlSupportedCustom()
Blend controls are disabled	Blend image is No image, closed, missing, or null	Select a valid equal-size B	validateParameters()
Preview area shows misaligned structure	Both images are cropped to the same pixel area, but B is not registered to A	Register and crop the pair first	process_utils::LeasedImage::read
Preserve lightness is disabled	Mode is Color/Lightness, or both inputs are grayscale	This is intentional; select another compatible RGB mode if needed	validateParameters()
Result is black where B is pure white in Color Dodge	Current exact-B=1 special case writes zero	Avoid exact white in B or use another dodge/lighten mode	an::BlendMode::ColorDodge
Colors/edges double or smear	Inputs share dimensions but are not registered/normalized	Register and match scale/background before blending	selector and worker validate dimensions only
Output intensity metadata looks implausible	Exposure metadata is summed independent of mode/opacity/mask	Treat combined exposure as implementation metadata, not photometric proof; correct downstream if needed	process_utils::combinedResultMetadata
OK with no B only stretches A, or appears to do nothing	Without a blend image the run applies Base Stretch alone; at its identity values it leaves A as it was	Select B before OK to blend; undo the step in the image's history if the stretch was not meant	ImageBlendProcess::executeTypedProcessImpl

Related workflow steps

Use Image Registration and normalization before blending data that must correspond pixel-for-pixel. Use Image Compare for non-destructive visual inspection, Starmask Recombination for its specific star/starless addition path, and Pixel Math for explicit formulas or output naming. Use Masks to prepare and inspect the active base mask.

References and provenance

This page documents AfterNight's current blend functions and focused tests. The equations are implementation contracts rather than claims of conformance to an external compositing standard.