Manual de Stacker

The adjustment matrix

When you do not know which engine or which settings suit a stack, the adjustment matrix finds out for you: it sweeps every ticked engine through a grid of fusion and finish settings on the baked stack, overnight if need be, and collects every output with its settings, metrics and contact-sheet crops. The Matrix view browses the results while the run continues.

The matrix is a tool, not a step: it has no card on the ribbon. You reach it from the File menu, and while a run exists the window shows a Matrix view in place of the usual step, with the run’s outputs listed at the left, a gallery in the centre and the run’s controls in the panel at the right. File ▸ Adjustment Matrix Results takes you back to that view from any step.

Three things make a matrix worth running. A new kind of subject, where you have no feel yet for which engine will win. A bracket that no single engine handles cleanly, where the best patch of each could be painted together on Touch up. And a quiet evening: the run needs nothing from you once it starts, and every output is browsable within minutes of being made.

Why. Stacker ships thirteen engines, and most of them have a knob or two that matter. Trying the combinations by hand means pressing Stack, waiting, looking, and keeping notes. The matrix presses the button for you, keeps the notes in a folder beside the project, measures each result the same way, and ranks them, so you spend your attention on looking rather than on bookkeeping.

Launching a run

File ▸ Adjustment Matrix… opens the launch dialog, titled Matrix run. The menu entry is available when the stack is baked, every image is developed, and nothing else is running; hovering a greyed entry reads Bake the stack first (Crop → Bake). The run folder sits beside the saved project file, so the project must be saved too: the dialog’s Start stays grey with Save the project first until it is. The panel at the right offers the same dialog as New adjustment matrix….

The dialog’s own description says what it does: Sweep every ticked engine through a grid of fusion and finish settings on this stack, overnight if need be. Every output lands in the run folder with its settings, metrics and contact-sheet crops; the Matrix stage browses them while the run continues. Below it, top to bottom:

Tier

The tier sets how wide the sweep is. It is the one choice that decides the run’s length.

Standard is the default. A deeper tier contains every variant of the tiers below it exactly once, and runs them first: a Deep run has its Quick results within minutes and its Standard results within the evening, so you can start browsing long before it ends. Changing the tier resets the Finish variants ticks to that tier’s defaults and gives the run a new name.

Tip. Run Quick first on any new kind of subject. Thirteen outputs is a number you can actually look at, and the ranking it produces reorders the engine picker on Stack so the likely winners are at the top. Come back for Standard once you know which two or three engines are in the running, and untick the rest.

Engines

One tick per engine. The ticks start as the engines you have ticked on Stack; if none are, all of them are. Every untick removes that engine’s whole share of the grid, so this is the quickest way to shorten a run.

Axes

An axis is one setting the run sweeps, with the values it will try. The table lists the axes of the ticked engines at the chosen tier, one row each: a tick with the axis’s name, the values, and the engines that sweep it. Hover a row’s tick to see the axis’s short key and the engines. Untick an axis to hold that setting at its base value for every engine, which shrinks the grid without dropping an engine.

At Quick the table says Quick runs each engine once at its preset — no axes. With no engine ticked it says Tick an engine to see its axes.

The Standard axes are the headline knobs: Levels, Tone match, Index smoothing, Focus metric, Patch preset, Adaptive consensus, Focus channels, and the engine-specific ones such as Max Contrast: enhanced, Depth Pro: smoothing radius, Selection radius or Studio Soft: focus prior. Deep widens several of those ranges and adds the secondary knobs: Depth prior, Slab frames, Halo suppression, Noise gate k, Consensus max planes, Min tile, Soft average, and more per-engine tunings. The Stack chapter explains what each of these settings does.

Finish variants

Each fusion can be finished more than one way without being fused again, and each way is one more output. The ticks are the finish variants available at the tier:

A fresh plan ticks the three basics up to Standard. At Deep the two extra stages come in and No finish drops out, because five raw fusions per Deep variant would be thousands of files nobody reviews. The dialog needs at least one variant ticked: with none, the estimate says Nothing to run — tick an engine and a finish variant.

Base settings

The axes change one setting at a time; everything else comes from the base. Project settings (hover: The axes are applied on top of this session’s Advanced, Patch and Finish settings.) uses what you have set up in this project. Built-in defaults (hover: A fresh project’s settings.) ignores them. Choose Project settings when you have already tuned the stack and want to explore around that tuning; choose Built-in defaults when you want an honest comparison of the engines as shipped.

Contact-sheet regions

Every output is scored and cropped at a few regions of the picture. Auto-pick lets the run choose them. Pinned uses the regions you pinned on the Stack view, and shows their count; it is greyed until you have pinned some, with the hint Pin regions on the Stack view first (Review ▸ Pin ROI, then drag).

Why. The automatic picks look for the kinds of place a stack goes wrong: fine detail, depth edges, flat areas. Your pinned regions tell the run what matters in this picture, an insect’s eye or a halo-prone edge, and both the scoring and the thumbnails then concentrate on exactly those spots.

Output

The run folder’s path is shown; it sits beside the saved project, or reads Save the project first — the run folder sits beside it in warning colour until the project is saved. Choose… picks a different folder.

Also write a 16-bit TIFF per output keeps a full-depth master of every output as well as the JPEG. The hover warns of the cost: ~120 MB each at 20 MP. The JPEG is always written. Leave it off unless you intend to print straight from a matrix output; a variant you like can be kept into the project later, at full depth, from its settings.

CPU only runs every engine on the CPU with no GPU accelerator. The hover gives the two reasons: for comparing against GPU results, or a machine whose GPU is needed elsewhere.

The estimate and Start

Above the buttons, one bold line sums the plan: so many outputs from so many fusions, an estimated time and an estimated size, for example 312 outputs from 104 fusions · ~2 h 41 min · ~3.1 GB. The time is based on how long each engine took the last time it ran in this session, or on a built-in table scaled by your image size and frame count when it has not run yet. Every change in the dialog recomputes it.

Start begins the run; while it is grey, its hover says why: Save the project first, Bake the stack first (Crop → Bake) or Wait for the running job. Cancel closes the dialog and remembers your choices for next time.

While the run runs

The status bar reads Matrix run and the run’s name started, the project is saved so that a crash mid-run leaves a resumable project, and the window switches to the Matrix view. The run’s status line at the top of the centre pane shows the run’s name, its tier as a chip, done / total outputs and a progress bar; the status bar’s job area at the right carries the running job’s progress and Cancel. The gallery and the strip fill in as outputs land, so you can browse, star and even keep outputs while the rest are still being made.

Outputs are scored as they finish. The run first makes the Quick tier, builds a pseudo-reference from those results (the pixel-by-pixel median across engines, so no one engine’s habits dominate), and scores every output against it at the contact-sheet regions. Five terms make the ranking:

Each term is converted to a z-score over the run, so that an output is judged against its siblings rather than on an absolute scale, then weighted and summed into one score; the highest score is rank #1. The ranking is a sorting aid, not a verdict: it reliably finds the clearly bad outputs and puts the plausible ones near the top, and your eye does the rest.

Two fusions with the same effective settings are run once; the second appears as a DUP marked output pointing at the first. An engine that fails on this stack is marked FAILED with its error in the detail view. When the run ends the status bar reads Matrix run finished — with the done, failed and duplicate counts and the elapsed time, and the run’s status line keeps Last run: with the same tally. If at least two engines were scored, the engine picker on Stack is reordered to the run’s best-first ranking; that order is remembered across projects.

Browsing the outputs

The strip at the left is headed Outputs with the count shown. Each row is one output: its thumbnail, the engine as a chip, its rank and score, the settings it swept, FAILED, DUP or KEPT badges, and a star that appears on hover. Click a row to open it. The ↑ and ↓ keys step through the rows with the zoom kept, so one spot can be compared across outputs at 100%.

The line under the header sums the view, for example 39 outputs · Rank ↓, or 38 of 39 · 2 hidden · Rank ↓ when something is filtered out; hover it for the whole filter in words. Clear on that line drops the filters and grouping and keeps the sort.

The gallery in the centre shows the same outputs as cards: the image, the engine chip, rank and score, the badges, a star (☆ / ★) and a Hide / Show toggle. Until the first output lands it reads Waiting for the first output…; with everything filtered out, Nothing matches the filter.

The filter

Filter on the strip’s header (hover: Sort, filter and group the outputs) opens the SORT & FILTER popover. While anything narrows the view the button reads Filter · with the count, and Reset at the popover’s top drops the filters and grouping, keeping the sort.

Under the rows: Show hidden outputs brings back what you hid, a search box (Search id or label) narrows by name, and the last line counts shown of total shown.

Tip. The fastest sweep through a Standard run is Sort › Rank, Group by › Engine, and Region set to your pinned must-be-sharp spot. Each engine’s best few are then at the top of its group, cropped to the place you care about, and ten seconds of scrolling tells you which two engines to look at properly.

The detail view

Click a row or a card to open the output large. The header has ← Gallery (or Esc) to go back, ◀ and ▶ for the previous and next output in the strip’s order (also ↑ and ↓), and, at the right, Folder (hover: Show this output’s files), Hide / Hidden, ☆ Star / ★ Starred, the output’s position (3 / 39), its rank chip (#1), any FAILED, DUPLICATE or KEPT chip, and its name.

Under the image, the comparison tools:

• A/B picks what to compare against: Off, the Stack step’s current result, every starred and kept output, the ten best-ranked, and this output’s finish siblings, each with its rank. With a side chosen, Hold to flip shows the other image while the mouse button is down on it, or hold X (Or hold X on hover). The zoom and position are shared, so the flip is the only thing that moves.
• Diff, offered when the other side is another output, shows the difference between the two instead: Luma absolute difference, brighter = more different. Where nothing lights up, the two settings made no difference and the simpler one will do.
• Depth shows the output’s depth map, the picture of which image was chosen where, when the engine produced one. Provenance shows which source image produced each pixel, when the output has that map.
• 1:1 under cursor (hover: Fetch a full-resolution window around the pointer.) loads the full-resolution pixels around the pointer, so you can judge fine detail without decoding the whole image. Fit (hover: Fit the whole output in view) brings a panned-away output back.

The panel at the right describes the open output in four sections:

Why. The Settings vs project table is the part to read before adopting anything. It is the exact list of changes that took the project’s result to this one, usually two or three lines, and often it shows that the difference you can see came from one knob.

Keeping what you like

Marks are cheap and are saved with the project; the Marks section in the panel counts them: so many starred, kept and hidden, with the note Stars, hidden outputs and kept variants are saved with the project.

• Star marks an output as a candidate. Stars make an output show up in every A/B list, keep it visible through Clean up…, and feed Keep all starred.
• Hide drops an output from the strip and the gallery; the summary line counts them and Filter › Show hidden outputs brings them back. Hide what you have ruled out, so each pass through the gallery is shorter than the last.
• Keep image (hover: Import this output as an extra result slot beside the engine results.) copies the output into the project as a stack of its own, marked KEPT, where Touch up can paint from it or onto it and Finish can export it. The status reads Keeping and the output’s name while it imports; an already kept output’s button reads Already kept on hover. Keep all starred at the bottom of the strip does the same for every starred output, one at a time, with Keeping n more… counting down.
• Apply settings (hover: Take this variant’s fusion and finish settings into the session, then press Stack.) copies the output’s settings into the project without copying its pixels. The status reads Settings applied from the output’s name — press Stack; any stack result those settings would change is marked stale. Use this when you want the winning settings at full quality and with your own engine options on the Stack step from here on.
• Save as subject… (hover: Keep this variant’s fusion settings as a starting point in Stack › Subject.) opens the save-subject dialog with the output’s fusion settings and thumbnail filled in. The session’s own settings stay as they are. Use it when this stack is one of a kind you will shoot again.

Tip. Keep the pixels when the output is the picture you want; apply the settings when you want to keep working on it. A kept variant is a finished fusion at the matrix’s quality and cannot be re-fused; applied settings give you the same result from a fresh Stack, with Touch up and every later step open.

Resuming, locating, cleaning up

A matrix run lives on disk, in its folder beside the project, and the project remembers it. Three things can happen to it.

Resuming. A run that stopped early, because you cancelled it or the computer went to sleep, can continue: Resume matrix run appears in the panel and on the run’s status line whenever outputs are missing, the run’s index is there, the stack is still baked and nothing else is running. Its hover reads Continue with the outputs not yet in the run’s index. Only the missing outputs are made; the finished ones are kept and scored exactly as before, because the reference the run scored against is saved in the folder too.

Locating and forgetting. If the run folder is moved, the status line reads Run folder not found: with the path, and the strip says Run folder not found — locate it or forget the run above. Locate… (hover: Point at the moved run folder (its plan.json must carry this run’s id).) reconnects it. Forget run (hover: Drop the record; kept variants stay in the project.) removes the run from the project; the status reads Matrix run forgotten — kept variants stay, and the window returns to Stack. If the folder has been deleted outright the line reads Run folder removed and only Forget run is offered. Open folder shows the run folder in your file manager; its hover gives the path.

Cleaning up. A Standard run is a few gigabytes and a Deep run many more. Clean up… in the panel, on the status line, or File ▸ Clean Up Adjustment Matrix… (hover: Delete what the run produced, keeping chosen variants first.) opens the Clean up adjustment matrix dialog. It first reads Measuring the run folder…, then states exactly what it will do:

• Will delete: the folder’s path and the tally, so many variant folders, fusions, contact-sheet pages, the reference/ folder and their size.
• Will keep: the kept variants and their size and the plan / index / ranking files, or No kept variants — only plan.json, the indexes and the rankings stay.
• If any starred outputs are not kept, an accent box lists them, n starred outputs are not kept and will be deleted, with Keep these first (hover: Import them as extra result slots, then come back here.). Click it, let the keeps finish, and the dialog is ready again with those outputs on the keep side.
• If the folder holds anything that is not part of the run’s layout, a warning lists it under Not part of the run’s layout, inside folders that go: and the delete button waits for I understand these will be deleted too. Entries at the top of the folder that do not belong to the run are never touched; the dialog counts them as other entries at the top of the folder are left in place.

Delete names the folder it will delete; Cancel closes the dialog with nothing removed, as does an error (Nothing was removed.). The status reads Cleaning up matrix run…, then Matrix run cleaned up — with the space freed. Afterwards the run’s status line carries a cleaned up chip, the strip’s summary reads kept only, and the kept outputs remain browsable; everything else is gone and Clean up… is no longer offered.

Why. The run folder is the evidence: every output with the settings that made it. Keeping chosen variants first and listing foreign files before deleting means a clean-up can never cost you the one output you meant to keep, or a file you dropped into the folder yourself.

Status messages on this step

The status bar’s left side reads ↑ / ↓ step through the outputs — the zoom stays put · click a row or card to open one · hold X to flip the A/B, with your own shortcuts filled in. The ribbon has no Matrix card; the gate that would read Run File › Adjustment Matrix… on a baked stack first is what the view waits for, and until a run exists the centre pane shows No adjustment matrix with File › Adjustment Matrix… sweeps every engine through a grid of settings on the baked stack and collects the results here., while the strip reads The outputs of a matrix run line up here, ready to step through.

Messages that appear as you work: