Stacker manual

Stack

Stack is where the sharp parts are chosen and combined. Everything before it made the images honest and lined them up; everything after it works on what Stack produces. The step asks three things of you: say what kind of picture this is, tick the engines you want to try, and press Stack. Then you judge what each engine made and keep the one you like.

Above them, a warning appears when the images about to be stacked mix file types (“The stack mixes file types: 21 ORF, 1 DNG.”) or when one of them is a result Stacker exported earlier, recognised by the program name it wrote into the file: stacked again, its detail would count twice. Take it out on the Images step unless that is what you want.

The panel has five sections, in the order you use them: Subject, Shared settings, Engines, Review and Export. Every section folds; the caption of a folded section carries small icons that say what is in force inside it, so you can read the state of the step without opening anything. Review and Export open on their own the first time a result arrives.

Until a result exists the view says No stack yet and Pick one or more engines in the panel, then press Stack. Once one exists, drag to pan and scroll to zoom, exactly as on Align.

Subject

A subject is a starting point: a set of values for the knobs every engine shares, chosen for one kind of picture. You pick the subject that matches what you photographed, and the engines start from values that suit it. Everything else on the step stays as it was.

Why. The same five knobs matter to every engine, and the right values depend more on the subject than on the engine. A hair-thin antenna wants small regions and little smoothing so the engines follow it; a coin wants large regions and heavy smoothing so it comes out as one even plane. Naming the subject once is quicker than finding those values in thirteen gears.

The starting point in effect

The dropdown at the top names the subject in effect: Starting point: followed by its name. The name is found two ways. If you picked a subject, that is the one. If you did not, the program looks at the knobs as they are now and names the subject they equal. When they equal none, the dropdown says Custom and its hover reads Your own settings; pick a starting point to reset them.

When you have picked a subject and then moved a knob in an engine’s gear, the dropdown adds · adjusted and its hover says adjusted in the engine gears. Your adjustment is in force; the subject is only the place you started. Pick the subject again to put its values back.

Under the dropdown a line reminds you of the division of labour: Starting points; fine-tune in each engine’s gear. When the section is folded, its caption shows the subject’s icon with the hover Starting point: and the name.

The built-in starting points

The three built-ins each set the same five knobs: the region size range the region engines grow, the index smoothing the depth engines apply, the number of pyramid levels, and the two radii Depth Pro measures and smooths with. They touch nothing else: slabs, the depth prior, tone matching and every other gear stay as you left them.

Each built-in has a one-line hint under its name in the list: Fine setae and antennae: small patches, tight index median, Smooth gradients: the balanced defaults, and Near-planar subjects: coarse patches, heavy index smoothing.

Tip. If you are not sure, start with Flower · smooth. It is the measured default, and the other two are departures from it in opposite directions.

The library and the guide

Click the dropdown to open the library. At the top is a search box, Search subjects…, that matches a subject’s name or its hint line. Under it three tabs narrow the list: All, Built-in and Custom. Then the rows: a tile, the name and the hint. Clicking a row picks that subject and closes the list. A saved subject’s row also carries two small buttons, Rename and Delete. At the bottom is Save current…, described below.

When the Custom tab is empty the list says No saved subjects yet. Save current… keeps every fusion setting and the ticked engines under a name. When a search finds nothing it says No subjects match and repeats what you typed.

Picking a subject changes the knobs, and any engine whose settings changed loses the result it had cached, so a picture made with the old values is never shown as if it were made with the new ones. Press Stack again to see the effect.

Detailed view… sits in the section’s caption, beside the ?, and opens the Subjects window. Help ▸ Subject Guide… opens the same window on its second tab. The window’s head says what it is for: Choose a starting point that matches your subject. and These are presets — you can always fine-tune any setting for your specific image. Two tabs at the right switch between Details and Key strengths guide.

Details has the same list down the left, with each row reading Show and the name, and the chosen subject’s pane on the right: its tile, its name and hint, and either a green In effect chip (with an Adjusted chip beside it when a gear has moved) or an Apply button. For a built-in the pane carries three cards:

• What it does: what the five knobs do for this subject, in one paragraph.
• Recommended use: where to reach for it.
• Key strengths: chips for the columns of the guide where this subject is rated Excellent.

For one of your own subjects the pane shows a Settings card instead, with what it holds: Engines, Patch size, Pyramid levels, Index smoothing, Depth prior, Depth radii and Gears moved (the engines whose own settings differ from the defaults, or None).

Key strengths guide is a table of the three built-ins against seven things a picture can ask for. Each column has a heading and a two-word caption; each cell is a rating of four pills. The subject in effect sits on a highlighted row.

A legend under the table spells out the four ratings, Excellent, Good, Fair and Not ideal, and a note repeats the point of the whole section: These are starting points. You can always fine-tune any subject for your specific image.

Saving your own

Save current…, at the bottom of the library, keeps the session as it is now under a name of your own. Its hover says exactly what is kept: Keeps every fusion setting and the ticked engines as they are now, under a name of your own. That is more than a built-in holds: a saved subject records every knob of every engine, the shared settings that the Stack button sends, and which engines were ticked, so picking it later brings all of that back at once. If the engine on screen has a result, a thumbnail of that result becomes the subject’s tile.

The Save current settings dialog asks for a Name; the field’s hint is What the picture is of. Save is enabled once the name is not blank, and Enter commits. The status bar then says Subject and the name saved, and the new subject is listed under Custom and becomes the starting point in effect.

Saved subjects can be renamed (Rename subject) and deleted (Delete subject). Deleting asks Delete “…”? The engines keep their current settings. A deleted subject that was the starting point leaves the dropdown on Custom; nothing about the engines changes. A variant you keep from an adjustment matrix run can be saved as a subject in the same way; the adjustment matrix chapter has that.

Tip. A saved subject is a Stack-stage thing. It does not keep the look, the stamp or anything from Finish, so one subject can serve pictures with very different finishes.

Shared settings

The knobs here are not any one engine’s. They shape the images every engine reads, or the pass every engine makes, and they are the only settings on the step that a built-in subject does not set.

When the section is folded, its caption shows three icons, lit when the thing is on: the depth prior (Depth prior on or Depth prior off), slabbing (Slabs of … images or Slabbing off) and lateral CA (Lateral CA correction on or off).

Focus-step depth prior

A camera that shoots its own focus bracket records, in each file, the focus step it was at. Focus-step depth prior uses those steps as a hint: it pulls each pixel’s choice toward the image the lens’s focus position says is nearest the pixel’s estimated depth. It affects Fine Detail, Edge Detail and Balanced.

It is off by default, and the hover says why: it can soften thin near edges such as hair and antennae. Try it only when the selection wanders on a featureless background, where no image is clearly sharpest and the engines pick noise. The prior needs a focus step in every image of the stack, and the steps must differ; files without that metadata give it nothing to work with.

Focus-step prior sigma is live while the prior is on. It sets the strictness of the lens-metadata depth prior: smaller is stricter. The range is 0.02 to 0.5 and the default 0.08.

The region engines, Sharp Regions and Adaptive, have their own switch for this prior inside their gear, and that switch is only read while this shared one is on.

Slabs for deep stacks

Slab size (images, 0 = off) is for very deep stacks. Instead of fusing all the images in one pass, the program fuses overlapping sub-stacks of this many images with the chosen engine, then merges the slab results by depth. That curbs the growth of noise across a long stack and bounds the memory a pass needs. The range is 0 to 64 and the default 0, off.

Slab overlap (images) is live once slabbing is on: how many images adjacent slabs share, so the seams between them blend. The range is 0 to 8; the default is 1.

Why. Most stacks never need slabs. Reach for them when a pass runs out of memory, or when a stack of a hundred or more images comes out visibly noisier than any single image in it.

Resampling

Resampling chooses the kernel the align and crop warp samples with, that is, how a pixel is computed when an image is shifted or scaled by a fraction of a pixel. The list runs from Spline 4×4 and Spline 6×6 through a family of Lanczos kernels, two Cubic presets and Nearest neighbour, to the fixed-point kernels: Spline 4×4 (fixed-point, GPU-exact), Catmull-Rom (fixed-point, GPU-exact) and Lanczos4 8×8 (fixed-point, GPU-exact). Spline 4×4 (fixed-point, GPU-exact) is the default for new sessions; Catmull-Rom (legacy) is what sessions saved before this setting existed used, and they keep it so their pixels do not change.

The choice takes effect on the next align, crop or stack. Images that are already baked keep the kernel they were baked with.

A status line under the picker says where the warp will run. With a fixed-point kernel it reads Warps and previews run on the GPU when one is available. With any other it warns No GPU shader for this kernel: warps and previews run on the CPU. The fixed-point Spline 4×4 looks the same and runs on the GPU.

Why. The fixed-point kernels give the same result bit for bit on the CPU and the GPU and on every platform. That is what lets a stack be reproduced exactly on another machine, which the science bundle promises.

Lateral CA correction and Re-develop

Lateral CA correction (develop) is the same switch as on Images › Camera develop, repeated here because it affects every engine. It removes the colour fringing a lens leaves toward the edges of the frame while the RAW is developed, before any engine sees the image. Off keeps the RAW’s own colour fringes. It is on by default.

The switch applies when images are developed, so changing it on an open stack does nothing until you press Re-develop. That button develops every source again with the current policy. Alignment, crop and touch-up strokes are kept; the stacks run again.

Why. A colour fringe is a sharp red or blue edge next to a real edge. An engine cannot tell it from detail and will stack it faithfully. Removing it at development is the only time it is cheap.

Your defaults

Save as my defaults makes new sessions start with the engine and shared settings as they are now. Reset goes back to the measured defaults for every engine and shared setting. It leaves alone what belongs to other steps: JPEG quality, the alignment settings and the develop settings. A line beside the two buttons says Applies to every engine.

A hint closes the section: Power, hardware and crash-report settings are in File › Settings. Nothing there changes a result.

Engines

An engine is one way of judging which image is sharpest at each point, and of combining the winners. Stacker ships thirteen. They do not all answer the same way, and no single one is best on every picture, so the list lets you tick several and run them in one press.

The first three rows sit under Recommended. They are ranked for your stacks: once you have run an adjustment matrix (the adjustment matrix chapter) the program scores the engines on your own pictures and keeps that order; before that it uses a measured order that puts clean, sharp engines first, then near-duplicates, then the grainy ones, then the soft ones. Show all engines lists the rest under More engines, grouped by family: Depth, Detail, Contrast, Soft and Blend. Show fewer engines folds them away again; an engine you ticked below the fold stays listed while the list is folded.

Each row is a checkbox and a gear. Tick the engines to run. Hover a name for the engine’s card: its name, what it produces, how it decides what to keep, its character, and where it ranks (Recommended for your stacks, or Ranked … of 13). When the section is folded, the caption shows how many engines are ticked and lights a gear with a dot if any engine’s settings have moved off the defaults.

The thirteen engines

What follows is each engine’s own card, in the order of the list before any ranking of your own. The families say which idea the engine is built on; the character is a fixed trait of its output.

Tip. Two engines that disagree are more useful than one. A classic pairing is a depth engine for the smooth surfaces and Max Contrast or Sharp Regions for the hairs; the Review section shows where they differ, and Touch up paints one onto the other. Touch up’s Add companion layer button sets up Depth Aware with Sharp Regions in one click.

An engine’s settings

The gear beside each engine opens a popover named after the engine, holding only the knobs that engine reads, with a Reset that puts back exactly those and nothing else. The gear’s hover says … settings, and … settings · changed with a dot on the gear when any of its knobs differ from the measured defaults. The gear is disabled while a job runs.

Some knobs are read by more than one engine. A line under such a knob says Shared with and names the others, so you know that moving it in one gear moves it in the other. The region engines share a whole set.

Changing a knob drops the cached result of every engine whose settings it touched. The engine’s thumbnail disappears from the result list until you press Stack again; a result fused with old values is never shown as current.

Fine Detail

Edge Detail

Depth Aware

Balanced

Sharp Regions and Adaptive

The two region engines share one body of settings; a change in either gear shows in the other.

Max Contrast

When Max Contrast or Fine Detail runs with tone matching in a press that also runs Depth Pro at its defaults, the depth pass is computed once and shared, so the extra pass costs nothing more.

Depth Pro

Soft Average

Depth Standard and Depth Legacy

Contrast Classic

Studio Soft

The alignment check

Stacking unaligned images smears every edge, so above the button the section says what state the images are in. Once Align has run, or the crop is baked, a green line reads … images aligned with the count. Otherwise an amber line reads Images are not aligned; its hover says Stacking unaligned images smears every edge. Run Align, or attest below if the images are already in register.

Images are already aligned is that attestation. Tick it when the images were registered before import, by a rail with no drift, by another program, or because they are scans. It is disabled once the images are aligned, because then it has nothing to say. The Align step card reads Skipped afterwards, with a hover that says what that costs.

Stack

Stack runs every ticked engine, one after another, one full pass over the stack each. It is enabled when every image has developed, there are at least two, nothing else is running, and the images are either aligned or attested. Pressing it with no engine ticked says Nothing to stack and Pick at least one stack engine.

If the images are aligned but the crop has not been baked, Stack bakes it first, with the status Baking the crop first so every engine reads warped images once: the same job the Crop button sends, so every engine then reads warped images from disk rather than warping the whole stack again on every pass. The automatic rectangle is used if you set none. If the aligned images share no common area at all the step refuses with Can’t stack these images.

While it runs the step card and the status bar show Stacking with the engine’s name and, for several engines, which pass this is of how many. As each engine finishes the status reads Stacked and its name; the last one is followed by Stack complete. A new stack clears the touch-up strokes and the finished picture, since both were made on the old result.

With Battery saver scheduling on and more than one engine ticked, a note under the button reminds you that Efficiency scheduling: each engine is a separate full pass over all … images.

The engine that finished last is the one on screen. Every engine’s result is kept, and they line up as cards on the Touch up step, where Use as base makes any of them the result the rest of the program works on.

Stack in background

Stack in background, under the Stack button and in the File menu, hands the bracket to a worker of its own and leaves the session free. You can browse, import the next bracket and start another background stack straight away; it waits behind the first. The images go in stack order, without the ones left out, with the ticked engines and the current look. The worker develops, aligns and stacks them again from the files, the way Batch… does, so alignment work done on the Align step is not carried over. Each engine’s 16-bit TIFF and its processing record land in your export folder; the first time, Stacker asks where that is. It is offered when there are at least two images, each read from a file, and an engine is ticked.

While a stack runs, the status bar shows a Background chip with its progress and how many are waiting. When one finishes, the status reads Background stack ready (or failed, with the reason), the chip stays lit until you look, and if Stacker is not in front its Dock or taskbar icon asks for attention. File ▸ Background Stacks… or the chip opens the list: Stop takes back a waiting stack or stops a running one at its next image, Reveal shows a result in the file manager, Open imports it as a new set the way a drop does, and Clear finished tidies the list. Quitting stops a running background stack instead of waiting for it.

Why stack again from the files: the background stack must not depend on the open session, which you are free to close, re-import or change while it runs.

Check

After every stack the Check section reads what the engine left behind and says, image by image, whether the stack would be better without it. The evidence is the engine’s own record of which image won each pixel, Align’s verdict on every image, and the exposure balance the engine applied; nothing is re-measured, so the check costs no time.

The findings, each naming the image:

• Won no pixels at the start or end. The bracket overshot the subject. Harmless; dropping those images only saves time.
• Won no pixels while its neighbours did. Soft, or a twin of the one before. The stack is sharper without it.
• Won a small fraction of what its neighbours did. Weak: usually a slightly missed frame.
• Could not be aligned and was left where it was. Align’s fallback; the image was bumped or has nothing to lock onto.
• Sits off the scale trend. Aligned, but its scale does not follow the rest: the camera or the subject was knocked between frames.
• Darker or brighter than the rest, in stops. A flash that did not fire, a cloud.

Restack without them (n) leaves the named images out, aligns the rest again and stacks with the same engines. The images stay in the list on Images, excluded, so any can be put back with the eye. The status line after a stack says how many images could go, so you hear about it even when the hands-off flow has moved on to Touch up.

The check needs at least four images and an engine that records which image won each pixel: Depth Aware, Fine Detail, Balanced, Sharp Regions, Adaptive, Depth Pro, the Soft engines and Contrast Classic do; Edge Detail and Max Contrast do not. When most of the bracket looks off, the check says so and suggests nothing: dropping images one at a time will not rescue a capture that needs reshooting.

Why advice and not an action: the check cannot know that the frame it would drop is the one with the antenna you wanted. It names the image and the reason; the click is yours.

Review

Tools for judging a result at 100 %. Every control is in place before a stack exists and enables as the results it needs arrive. When the section is folded, its caption shows how many regions are pinned and lights an eye when any overlay is on.

Pinning a region

Pin ROI is a checkbox that turns the view into a marking tool. With it on, drag on the stack to draw a rectangle. A Mark region dialog then shows the region’s size and position (… px at (…, …) in the result), a row of tags for what you see there (must be sharp, halo, ghost, soft, motion, ok), a note field whose hint is what you see here, and Save or Cancel. Pinned regions are drawn over the result. A count beside the checkbox reads … pinned, and the × at the right, whose hover is Remove every pinned region, clears them all.

Pinned regions become the adjustment matrix’s contact-sheet crops: when the matrix runs, every variant is cropped to your pinned regions so you compare the places you care about. The adjustment matrix chapter has the matrix.

Tip. Pin the places a stack is most likely to get wrong: a hair crossing a far background, a bright edge in front of something dark, a smooth surface that should have no texture.

The depth map

Show depth map replaces the result with the engine’s depth map: which image won at each point, as a grey ramp from black for the first image to white for the last. It is enabled when the engine on screen produced one; not every engine does. Save depth map… writes it as a 16-bit grayscale TIFF: 0 = first image, white = last image. A depth map is data, not a picture, so the free-mode export mark is never applied to it.

The overlays

Three overlays paint over the result where each pixel came from or what happened to it. Each enables once the engine on screen has something to show.

• Low-confidence overlay: red where the engine’s selection was contested: the margin between the best image and the second was thin, or the pixel sits in the occlusion band around a deep depth step. Review attention, not error. Enabled when the engine produced a confidence map.
• Clipping: red where a channel clipped at white, blue where all channels are at black. Max Contrast can push values past the range before the flatten.
• Disagreement overlay: bright where two engines disagree, once a comparison has been made (below).

Comparing two engines

Compare with lists every other engine that has a result of the same stack, and Compare computes where that result and the one on screen differ. The map is painted over the result as the Disagreement overlay: bright where the two engines disagree about the same pixels, which are the places worth reviewing at 100 %. The status bar says Disagreement between the two names — bright = review attention. Where the two agree there is nothing to look at; where they do not, one of them is wrong, and the next step paints the right one over.

If the two results are not the same size the program asks you to restack both; if one has no cached result it asks you to tick it and press Stack.

Where a pixel came from

Most engines record which image won at each pixel. When the engine on screen did, a line under the overlays says Hover the image to see each pixel’s source image., and the status bar names the file under the pointer: pixel from the file name and, when the file carries it, its focus step. That is the quickest way to check that a hair came from the image where it was actually sharp.

Export

The Stack step can export directly, without going through Finish. The buttons are the same ones Finish offers; the Finish chapter has them in full, and the notes here are what you need on this step.

The formats

Each button opens that format’s settings first, then asks where to save. A line under the buttons sums them up: TIFF carries camera tags + a JSON processing record · JPEG and AVIF for sharing · PNG for lossless 8/16-bit. Each button opens its settings.

Every settings dialog is titled Export and the format, repeats the line above, and ends with Choose file…, Cancel and a Reset that puts the format’s defaults back.

Under the format’s settings, Also save as lists the other formats with a tick each. Tick JPEG in the TIFF dialog, say, and one Choose file… writes both: the TIFF where you pointed, and a JPEG beside it with the same name, using the JPEG settings as you last left them in its own dialog. The ticks are remembered with the export settings, so a stack that always goes out as TIFF plus JPEG plus DNG is one dialog and one file name from then on. The status bar names each file as it lands. The save dialog opens in the folder the last export went to, unless a host program’s hand-off names one.

What every file carries

Whatever the format, the file carries the camera’s facts as EXIF, read from the reference image when the stack was made: make and model, the lens, the capture date and time, the shutter speed, the f-number, ISO, the focal length, and the GPS fix when the camera recorded one; with the photographer and copyright from the stamp, the software, and a description naming the number of images and the engine. A JPEG holds it in its EXIF segment, a PNG in its eXIf chunk, an AVIF in its EXIF item, and a TIFF or DNG as tags of its own, so a DAM, a photo site or exiftool sees the stack as a photograph taken on that camera. The TIFF also carries an XMP packet with the creator, the rights, the capture date, and under SpikerSoft’s own namespace the number of images stacked and the engine; the DNG’s packet is Lightroom’s own, as the DNG section describes. The share set’s slides carry the same EXIF. The processing record is the separate JSON sidecar, written with a TIFF when its option is on and with every DNG.

• TIFF: Depth (16-bit, the default, or 8-bit), Compression (None, LZW or Deflate) and Camera tags + processing record (.json sidecar), on by default. The hint: 16-bit uncompressed is the archival default. LZW and Deflate are lossless; 8-bit halves the file with a dithered conversion.
• JPEG: Quality, the same slider as the panel’s. 92 is visually lossless for most macro work. This is the same quality the share set, batch runs and publishing use.
• PNG: Depth and Compression (Fast, Balanced or Smallest). Always lossless — compression only trades time for bytes. 16-bit keeps every level the stack produced.
• AVIF: Quality (30 to 100, default 80), Speed (1 to 10, default 4) and 10-bit coding (smoother gradients). 80 looks like a quality-92 JPEG at roughly half the bytes. Lower speed = smaller file, longer encode.

TIFF and PNG also ask which Colour the file holds. Display sRGB is the photograph as you see it: profile, look and stamp applied, sRGB for print and screen. Rec.2020 (scene) is the pipeline’s own pixels for an editor: scene-referred, no profile, look or stamp, Rec.2020 primaries with white at full scale (light above white is clipped), with the Rec.2020 ICC profile embedded. JPEG and AVIF are always display sRGB; their dialog says so and points you to TIFF or PNG for the pipeline’s pixels.

Pan video

Pan video… makes a slow scroll across the full-resolution plate, framed for Instagram, as an H.264 MP4. The Export pan video dialog explains the idea: the full plate scrolls slowly through an Instagram-sized frame; the photo is resampled once at the frame’s height, so fine detail stays sharp, and the clip is H.264 in an MP4 Instagram accepts as-is.

A summary line gives the frame size, the frame rate (30 fps) and the number of frames the clip will have. Choose file… asks where to save the MP4, and the status bar shows Encoding pan video with the frame count while it renders.

JPEG quality

JPEG quality sets the quality of JPEG exports and share tiles, from 70 to 100. It is the same value the JPEG dialog, the share set, batch runs and publishing all read, so there is one JPEG quality everywhere.

Science bundle

Science bundle… is an evidence-grade export of the current base stack. It asks for a folder (Folder for the science bundle) and writes, named after the engine:

• the fused TIFF exactly as the engine produced it: no touch-up, no finish, no display render, in the pipeline’s working colour space with the matching profile, and, when that space has headroom, the same pixels again as 32-bit float linear light;
• a provenance raster: a 16-bit grey image whose value at each pixel is the index of the source image it was reconstructed from;
• a depth raster in millimetres, when at least two images report a lens focus distance: each pixel’s value is the distance of the image it came from, quantised to the bracket’s step and only as accurate as the lens’s own readout;
• a confidence raster, when the engine produced one;
• a processing record, a JSON file with every setting, the artifacts above and a SHA-256 hash of every source image, so anyone can check that the files you hand them are the files the stack was made from.

The step card shows Science bundle while it writes, and the status bar says Exported and the folder when it is done. The bundle is never rendered through the look: it is evidence, and it stays scene-referred so a comparison with another program’s output means what it says.

Why. A science export skips every creative decision on purpose. If you need to show that a structure in the picture is real, the provenance raster says which photograph it came from and the hash says that photograph is unchanged.

Status messages on this step

The status bar’s standing hint on this step is Pick engines in the panel, then Stack · drag to pan · scroll to zoom. The step card has no idle note of its own; it reads Stack and, once a result exists, shows the step as done.

The step is dark until two images have developed; its hover then reads Import at least two images first. When Align or Crop was skipped, their cards say Skipped — the images are stacked unaligned; Align first for a sharper stack and Skipped — Stack will crop to the automatic rectangle.

While something runs, the step card and the status bar carry the job’s label, its count and an estimate of the time left:

Errors open a dialog rather than a status line: Nothing to stack (no engine ticked), Can’t stack these images (the aligned images share no common area), and the comparison’s stacks have different sizes — restack both engines or both engines need a cached stack — check them and press Stack.