Contents

What is in here

The chapters follow the order the app works in, so this can sit open beside it. Every number here is one the app measures, and where a threshold has a reason the reason is given — because that is the part that tells you what to do when a roll does not behave.

Chapter one

What this is

A colour negative is three dye layers. Trichroma Lab measures them one at a time, under three narrow bands of light, and prints the result through a pipeline of the kind a film lab runs.

Photograph a negative under white light and your sensor measures those dyes through its own red, green and blue filters. Those filters are wide and they overlap: the camera's red channel sees some of what the magenta dye is doing, the green channel sees some of the cyan. Each layer's measurement is contaminated by the other two, and the orange mask has to be guessed away afterwards.

Shoot the same frame three times instead, once under each narrow band, and every exposure measures one dye layer and almost nothing else.

Three exposures per frame go in and one finished positive comes out. Between them the app measures the light, measures the film base, registers the three channels onto each other, and grades the roll.

White light is supported too

One exposure per frame under a white lamp is a supported way in. It is harder, not impossible: under white light the sensor's own filters are half the instrument, so that mode is calibrated for a named station rather than offered blind. Everything in this manual applies to both unless a section says otherwise.

What the app does not do

It does not decode your raws into a picture and then filter it. It does not guess your film stock from the image. It does not adjust per frame what it can measure once per roll. And it never moves or changes the raws you import: a roll keeps its own copy of the sensor data it needs, so once PROCESS has finished the raws can go to your archive.

Chapter two

The rig

A camera on a copy stand, a film holder, and a lamp that can put out three narrow bands one at a time. Once that is built and measured, it stays measured.

The lamp

The rig this manual is written against uses a Scanlight v4 — an open-source RGB scanning light by jackw01, driven from a browser-based control app. It is not our hardware and we do not sell it. It is named here because it is what the measurements in this manual were taken on, and because its control app is where the numbers in the next chapter are typed.

Any lamp that can produce three sufficiently narrow bands will do. What the app needs from it is not a brand: it is that each band is narrow, that the lamp holds still while a roll is shot, and that you can write down where the knob was.

The camera and the stand

Shoot raw. Fix the camera and the film in place and do not move either between the three exposures — the app corrects small residual drift, and does not stand in for a tripod. Turn in-body stabilisation off as well: on a stand it has nothing to steady, and it can move the sensor between exposures. And shoot single frames rather than a burst: some cameras write fewer bits per photosite in continuous drive. On a Nikon, choose lossless compressed NEF: the High Efficiency formats cannot be read.

One shutter, one ISO

Manual exposure throughout, and the same shutter and ISO across the three bands of any one set. This matters more than it looks. A calibration split across two shutter speeds carries about 90 counts of error into every density on the roll while looking perfectly ordinary, which is why the app refuses to save one.

The holder

Whatever holds the film flat, as long as two things are true when you look through the camera: the frame and the clear film around it are both visible, and the black holder can be left outside the box you will draw in chapter eight.

That clear film around the frame — the rebate — is not spare margin. It is unexposed film base, and it is where the roll's own base density is read. Chapter eight is about protecting it.

What gets measured once

The first three are the rig. The last two belong to the film and are measured again for every roll you shoot, because they are properties of the film and not of the bench.

Chapter three

Setting the lamp

Take the film out. Point the camera at the empty gate. Set each band so the sensor is filled to between 75 and 90 % of its range, with nothing at the ceiling — then write the three numbers down.

This is the one part of the process that happens outside the app, in the lamp's own control software. What you are looking for is not a pretty picture: it is three exposures of nothing at all, each one as bright as it can be without touching the top.

Why 75 to 90 %

Below 75 %, the flat wastes well capacity, and with it the signal-to-noise of every density measured against it. The flat is a divisor: noise in it becomes noise in every frame on the roll.

The top of the window is 90 % and not 100 %, and the gap is not caution — it is drift. A lamp was measured moving 1.3 % between two consecutive exposures at one setting, and a white lamp measured falling 7.2 % across a single sitting with the exposure unchanged. A flat set at 96 % is not clipped today and is clipped next week.

Working towards it

You will not hit the window first try, and you are not meant to. Type a setting, shoot the gate, drop the file into the app, and it will tell you where that setting landed and suggest the next one. Two or three rounds is normal.

The app suggests rather than calculates because no formula survives contact with somebody else's hardware — the three channels of a real lamp do not share a response, a span, or a starting point. It learns yours from the settings you actually tried.

If you are using a white lamp

The window is different, and so is the instruction. One broad capture read through the camera's colour filters cannot put all three channels into 75–90 %, because green collects roughly twice what red and blue do. Aiming a white exposure at that window teaches you to blow out your green.

For white there is one exposure and one rule: stay clear of the ceiling. The app wants it under 92 % and not below 60 %, and says which way to move.

Automation

If your lamp's control app can fire the camera, use it. Three exposures triggered in sequence with the light changing between them is three identically framed exposures; three exposures triggered by hand is three chances to nudge the stand. Keeping the light on between frames also avoids warm-up drift inside a roll.

Chapter four

Flats — the denominator

A density is a ratio. The flat is the bottom of it.

Every density in the scan is measured against the empty gate:

D = −log₁₀ ( frame ÷ flat )

So the flat is not a nicety of the capture. It is the denominator of the whole measurement, and a flat that is clipped at the top pulls every density in that channel down without leaving a mark on the rendered picture.

Put the other way round: a roll with no flat does not have worse densities. It has none — every number downstream is a fraction of the sensor rather than of the light.

Flats are taken out of the roll

Include the three bare-light exposures with the roll when you import, and the app finds them and lifts them out. They are the light, not frames.

They come out before grouping, and that ordering is deliberate: three bare-light exposures shot seconds apart are a perfect frame as far as the grouper is concerned — same sequence, one per band. A set left in the pile becomes a phantom frame that gets graded, counted and exported like any other.

If you have neither

The app will not stop you, but it will not pretend either. You can waive the light, and then the roll is anchored on a film-stock table instead of on your own measurement, its densities come out in synthetic units, and the save button says so: process N frames · no light.

One bare-light frame anywhere in the roll retires the waiver by itself. The app finds it.

Chapter five

Calibrating the rig

Three shots of the empty gate, one under each diode — the denominator of the last chapter, measured and written down.

Order does not matter, names do not matter

You cannot put the red exposure in the green well by mistake. The app decides which band a file is by looking at it — the largest median names it — and the reason is blunt: the file name is not evidence. It is typed by hand and it has been wrong. A red exposure of the empty gate reads something like 12474 / 1683 / 517 across the three channels; nothing else looks like that.

What the sheet tells you

Six verdicts are possible. They are checked in this order, and each one says what is wrong rather than just that something is:

Only the first three have to be shot again. The rest are advice.

Uniformity

Beside the level sits a second number — uniform 0.95 or similar. It is the middle of the exposure divided by its brightest part, and it is what tells a flat from a frame. A real flat reads around 0.92 to 0.94; a frame with a picture on it reads 0.08 to 0.26. Anything under 0.70 is rejected as not a flat at all.

That check exists because the level alone would wave a frame through: the brightest part of a frame is its rebate — bare film — which passes nearly as much light as the empty gate. On level alone, a negative looks like a perfectly reasonable flat.

Saving does not require GOOD

This surprises people, so it is worth stating plainly. The app will save a calibration whose bands are outside the window. What it insists on is different:

• All three bands present
• All three actually flats
• All three shot at the same shutter and ISO

A dim flat is a true picture of a dim light, and refusing to record it would leave the roll with no denominator at all. The screen says what is not ideal; the gate refuses only what is not usable.

Save to rig, or use for this roll

The same sheet does two different jobs, and which one it is doing is written on its button.

A roll's flats are a measurement of one session, and a session is not a calibration — so they are never written to the rig. This distinction was once a single button, and it re-calibrated the whole station from three exposures shot beside one roll while leaving that roll with no flat at all.

Which flat a roll uses

Shooting flats with every roll costs three frames and removes an entire class of silent error. It is the single cheapest habit in this whole process.

Two lamps are kept apart

If you use both a narrowband lamp and a white one, the rig holds two separate calibrations and the app always knows which is which. This is not tidiness. A white roll divided by the narrowband flat reads three diodes under a different lamp, and the base spacings it then reports are wrong in a way nothing downstream can detect — measured at 0.48 where the film's own spacing is 0.22.

A calibration has a date on it

When the sheet opens on a calibration you saved earlier, it says so and shows the date. A lamp drifts, so a screen that showed last year's calibration as though it were current would be worse than one that showed nothing. Drop new exposures onto a well to replace a band.

Falloff

The flats carry more than a level. From them the app builds a map of how your lamp and lens fall off towards the corners, and divides it out of every measurement and every rendered frame. Without it, that falloff lands in the film base as invented density — measured at 7–19 counts on this rig, and 20 counts of green costs up to +2.11 dE.

It is all three bands or none: a partial correction would tilt the base's red-to-blue balance, which is worse than no correction at all. There is no per-image switch, because the field is a property of the light.

Chapter six

Shooting a roll

Three exposures per frame, in a consistent order, without moving anything. The app will work out the rest.

You do not have to name anything

Drop the whole folder in. Which three exposures belong to one frame is worked out from filenames, capture time and shot order — and from the pictures themselves, because a red-filtered negative measures nothing like a blue one. DSCF0001.RAF through DSCF0042.RAF straight off the card is a roll the app can read.

Which band a file is comes from the image alone, and the name is never trusted for it. A name is typed by hand and has been wrong; under one narrow band a single colour of the array dominates by a wide margin, and the largest median names it.

If you would rather make it certain

• One folder per channel, or _R, _G, _B in the file name — the strongest hint
• Shot in R, G, B order — capture time is read as well

Neither is required. They only shorten the work the grouper has to do.

Include the flats

Shoot the three bare-gate exposures at the head or the tail of the roll, with the film out of the holder and the light and camera on the same settings as the frames. Import them along with everything else.

Do not move the rig mid-roll

Registration is measured and corrected per frame, but it is correcting small residual drift. If the stand is knocked, the frames before and after are two different rigs sharing one flat.

Chapter seven

Merge

Drop the raws in. MERGE groups them into frames, lifts out the flats, and shows you what it decided before anything is processed.

Importing adds to the roll — it does not replace it, so one roll's files can come in from several folders in several goes. Keep it to one roll of film: the roll is graded from one base, and another roll's frames pull that base toward a film neither of them is. Importing copies or moves nothing — the roll's own copy of the sensor data is made at PROCESS.

What to check before processing

• The frame count. Whatever you dropped in, minus the flats, should divide by three. In figure 7.2 that is forty-five in, three flats out, forty-two left, fourteen frames. One frame too many usually means a flat is still in the roll.
• The flats panel. It should name three files and show their levels — 90, 89 and 89 % in figure 7.2. Note that the rig above it says never calibrated and the roll is still fine: these flats were shot beside it, and a session flat outranks a rig one anyway.
• The separations. Click a frame and look at its three exposures. They should be recognisably the same picture in three different densities.
• Anything unmatched. Files the grouper could not place are listed rather than quietly dropped.

Grouped by capture time

The header says which rule was used. Where filenames carry channel markers the app prefers those; otherwise it falls back to capture time and shot order, and confirms each group against the images themselves.

Chapter eight

The base area

One box, drawn once per holder, that tells the app where the film is inside the capture — and where to read the roll's own base density.

The instruction on the sheet is the whole chapter in two sentences: mark the negative and the clear border around it. That border is unexposed film base — it is where we read the roll's base density, so it has to be inside the box. Leave the black holder outside.

The checklist

• The image
• The clear border all the way around
• Any part of the black holder

Being generous outward is the dangerous direction. Being generous inward only samples less base, which costs precision rather than truth — so when in doubt, pull the box in.

You do this once

The box is saved with the rig, so every roll you shoot on that setup arrives already set. The holder buttons — 35 mm, 6×4.5, 6×6, 6×7, 6×9, 4×5, Free — are slots, not shapes: 35 mm means the box I measured for my 35 mm holder, not a 3:2 rectangle. Nothing constrains the aspect, and each slot remembers the last box you left in it.

The ring

The ring is the band between the box's edge and the picture — the part that is read as base. Its width is set as a fraction of the frame, which is why the sheet also prints it in pixels:

RING 363 × 207 px

Pixels are the one unit that means the same thing on every holder. The fraction does not: 0.08 on 35 mm is about 1.1 mm of film, while 0.06 on 6×7 is about 1.8 mm. Anybody reasoning in fractions raises the number for the bigger holder and drives the ring into the picture.

Auto crop, and holding a ratio

AUTO CROP takes the rebate off every frame as the roll is processed, adjustable afterwards. It decides whether the crop is applied, not whether it is measured — every frame is measured either way, which is what makes turning it on later instant.

HOLD 3:2 and its equivalents lock the per-frame crop to the holder's own ratio. It never applies to the base area itself: the base area stays whatever you measured for the window, rebate and all, because the ring is where the base is read.

Chapter nine

The roll file

A roll is one file. Everything the session decides lives in it, so the roll can be reopened, moved between machines, and re-rendered from scratch.

What goes in it

• The rig calibration — the station, its response curves and this crop, as one object
• Every frame's D, C, M and Y, its tone and its crop
• The sensor data of every frame — a copy of what the roll needs, not your raw files

The raws you import are never moved or changed, and the roll does not read them again after PROCESS: it reopens, re-renders and exports from its own copy. Keep the raws as your archive — processing again starts from them. The copy keeps at least two bytes for every photosite, whatever the raw's own compression, so a roll can be as large as the raws it was made from, or larger; TRIM OUTSIDE BASE AREA, on step 1 of SET UP THE ROLL, keeps only the base area — the box — and makes a new roll smaller. Everything you export — TIFF, JPEG, NEGATIVE — is an open format you can take anywhere.

Because the calibration travels inside the roll, a roll opened on another machine renders identically there. It does not go looking for that machine's rig.

Chapter ten

Prescan — what it measures

Before a single frame is graded, the roll is measured: its film base, per channel, and how far the three separations sat from each other.

The film base

Three densities, always in this order and always increasing:

That rising order is the orange mask, and the app checks for it: three densities that are not in mask order are refused, because they cannot be a colour negative's base.

How the base is found

Three ways, tried in a measured order.

The lines go first, and the reason is that film is advanced by hand. One box is set per holder and every frame is read through it, but the frame that box fits at the start of the roll has moved by the end — so the ring reads whatever happens to lie at its edge. Measured across one roll, the ring spread 459–497 counts where the lines spread 13–16.

The ring is the rescue, not the disgrace: at small preview sizes the lines have too little to work with, and the ring carries the roll. The panel always says which one owned the reading.

One estimator owns the roll

Never a mean across two. Averaging a good estimator with a poor one produces a number that belongs to neither, so the app picks one and says so — “by the lines over 12 frames (2 by the ring, not averaged)”.

The spread is the measurement

Film base is physically constant within a roll. Every difference between frames is therefore error, not variation — which is why the app reports the spread and calls the frames agreeing only when every channel is within 10 counts.

Ten is not a noise floor. It is a price: 20 counts of error costs up to +2.11 dE, and over-reading costs roughly twice what under-reading does.

Registration

Two estimates per frame — green against red, and blue against red — and the line reports the median, the worst, and how many were refused:

registration 1.68 px median · 2.26 px worst · 4 refused

The shift is measured in fractions of a pixel and resampled rather than rounded to the nearest whole one. An estimate is refused when the frame has too little texture to register, when the shift is larger than 2 % of the width — that is a different picture, not drift — or when the best match is too poor to trust. A refusal changes nothing: the app leaves the pixels alone and says why.

The two charts

The cross-sections and the surface are two views of your light across the frame, averaged over the roll's frames. Lines that never flatten out are the sign worth knowing: they mean the base area is sitting on the picture rather than on clear film.

Chapter eleven

Grading

LAB is where the roll is printed. The vocabulary is a lab's: four axes, applied to the whole roll or to one frame.

The four axes

Roll correction and frame correction

The same four axes appear twice, and the difference is what they apply to. The left rail is the whole roll. The right rail is the frame you are looking at. A frame renders as its own numbers plus the roll's.

Grade the roll first and the frames second. The roll-wide axes are the ground everything else is judged against, and they render live — nothing is re-measured when you move them.

Base trim

Base trim is different in kind, and the app treats it differently. It shifts the measured film base itself, equally on all three channels, in counts — two counts a step, up to about forty either way.

Because it changes the base, it changes what every law downstream reads, so it does not take effect as you drag. Hold PREVIEW to see the frame as it would land; press APPLY to re-render the roll.

It is a neutral shift, not a colour axis — which is what keeps the mask order a real base has. There is no base trim on a white roll, and none on a roll whose base was refused.

Roll contrast

Four positions: −3 −2 −1 0. Zero is the roll as its own laws print it, which is why there is nothing above zero — the ladder runs from what the app already prints down to as flat as the profiles go.

Grey point

The pipette arms the next click on the picture and goes out with it — one shot, not a mode. It moves C, M and Y only, never D, and it lands on whichever set of rows it was started from: click it in the left rail and it moves the roll, click it in the right rail and it moves the frame.

If there is nothing neutral where you clicked, it says no neutral reading there and changes nothing.

Tone

Seven named looks, shown as live thumbnails of the frame you are on: hover previews, a press keeps. You pick a look, not a curve.

The two ends of the tone range move independently and their regions do not overlap, so a combined profile is exactly the sum of the two single-ended ones. Hard opens the separation — shadows down, highlights up; soft closes it. Alongside them sit a HYPERTONE switch and a strength axis.

BEFORE

Press and hold BEFORE to drop back to neutral. It takes off this frame's rows, never the roll's — the roll correction and roll contrast are the ground the frame is being judged against, so stripping them too would answer a question nobody asked and make every frame look far more corrected than it is. What BEFORE means is: this frame, before I touched it.

The same press-and-hold is the base trim's PREVIEW, deliberately, because it is the same question — what does the other version look like?

Copy, paste, sync

SYNC gives every other frame on the strip this frame's colour, tone and image corrections, replacing their own. Crops, turns and the trash are left alone, and it asks first.

The filmstrip

Each thumbnail carries its own four numbers — D0 C+1 M0 Y0 — so the state of the whole roll is readable at a glance without selecting anything. Frames sent to the trash leave the count but stay recoverable; the header shows the roll's frames and how many are in the trash.

Chapter twelve

Settings

Almost nothing in here changes what the app measures. It changes how the app answers your hand — how far one step moves, which way the density bar points, and how the preview is drawn.

What one step is worth

The bars count whole steps — D and C, M, Y to ±12 — and this is what one of those steps is worth. The frame's own rows and ROLL CORRECTION each carry their own unit, adjustable from 0.1 to 3.0 in tenths.

With a roll open these belong to that roll: the frame re-develops as you type, the roll keeps them, and they are what the next new roll starts from.

Which way +D points

The choice chapter eleven mentioned lives here: +D can mean brighter — the default — or darker, the way a printer reads it.

The correction keys

Q/A, W/S and E/D step C, M and Y from anywhere on the LAB screen; Z, X and C are D−, neutral and D+, and follow whichever way the density bar is pointing. The setting decides whether the top row of that keypad adds or subtracts — on a minilab keypad it takes a dye out, which moves the picture toward red, green and blue.

The live preview

The canvas can be drawn by the processor or by the graphics card. The graphics card updates the picture much faster while you adjust it. With SCANNER, its colours can differ from the file by up to two levels in 255.

The file does not depend on this choice: the export always uses the processor.

The colour chain

16-BIT, the default, keeps the fractions the scanner's chain rounds away at ten bits: smoother deep shadows under a strong push, and TIFFs written at sixteen bits with about ten and a half real ones. SCANNER writes eight-bit TIFFs. The picture keeps the scanner's level on either setting.

Chapter thirteen

Export

Three formats. The shape of that list is an honest statement about the machine rather than a product decision.

A TIFF positive is sixteen bits by default. That is not sixteen measured bits: the colour chain keeps about ten and a half real bits of tone, and an eight-bit file would round them away. SCANNER — the colour chain on the PREVIEW tab of SETTINGS — rounds at ten bits and writes eight-bit TIFFs. COMPRESSION on the export sheet is lossless either way.

The negative carries no profile, on purpose

It is not in a colour space. It is transmittance counts off a sensor with a filter in front of it, and stamping sRGB on that would tell every converter downstream to apply a transfer function to numbers that have never had one. This is the export to take into another converter.

Size is a develop width, not a resize

FULL, 4096, 2048 or a number you type. The decoder reconstructs straight to the size asked for, so a smaller export is genuinely cheaper rather than a big one scaled down afterwards. The custom box is in pixels on the long edge, because a frame on its end is taller than it is wide.

There is no size row for the negative: the measurement has no size to choose.

Cancelling

Files already written stay on disk. A cancelled export is a shorter export, not a failed one — half a folder of TIFFs is half a folder of TIFFs.

Chapter fourteen

Do and do not

Every line below is a mistake that has actually been made on this bench, with what it cost.

Chapter fifteen

When it refuses

A refusal always carries a sentence. The app is designed so that a frame arrives either with a measurement or with a reason — never with a number it does not believe.

A whole roll refused

If every frame refuses as denser than any film base, and the readings agree with each other, the film itself is probably fogged. The app recognises this: when the readings are consistent, still in mask order, and only past the window rather than absurd, it refuses the base outright and grades the roll without one.

That is not a workaround. It is what a lab scanner does with the same film, measured against its own logs. The panel says so: refused like the scanner · graded without a base. Base trim is withdrawn on such a roll, because the laws read nothing there and a control that did nothing would teach you that the panel lies.

When the numbers agree and are still wrong

Agreement across frames cannot tell fog apart from three other things that are also constant on every frame: mask penumbra inside the crop, holder in the corners of a strip that is not quite square, or a roll of one flat dense scene. No cheap statistic separates them.

What protects you there is the base area itself, drawn where the light actually is, and the cross-section chart on the prescan screen — lines that never flatten mean the box is on the picture.