The whole idea, first. A colour negative was made to be printed on paper, and the paper supplied most of the contrast. Scanning skips the paper, so that contrast has to be put back, and that is what the resting midtone does. The rest of this page is where the handles sit, and why they read in the film's own units.
Levels on a negative are not the same instrument as levels on a photograph. On a photograph you are adjusting a picture. On a negative you are placing landmarks on a measurement: saying where the film's own black sits, where its white sits, and how the range between them should be read. The interface should let you do that in the units the measurement is in.
That interface belongs to Open Negative Initiative, a free macOS application for converting and editing scanned film: colour negatives, positives and black & white, from a camera scan or a scanner. It runs on its own, with no host application to buy. Every frame opens with these handles at rest. macOS 13 or later. Apple Silicon and Intel.
Free, macOS 13 or later. Every frame opens at rest: nothing is applied until you move a handle.
The axis is a density, and it is not 0 to 1
The three per-channel sets are graduated from −0.35 to 3.0, in density. Those two numbers are not decoration:
- The top, 3.0, is the resting white, roughly the maximum density a colour negative reaches. It is where a channel's dense end lands when nothing has been dialled.
- The bottom is below zero, and it has to be. A density of zero means a channel transmitting all the light; a scan can and does contain values transmitting more than the reference, which arrive as negative densities. One of the reference films sits at −0.243 in density, and the track runs down to −0.35 to leave room under the worst of them. A track stopping at zero would put those pixels out of reach of the black point.
That is one of two reasons a scan holds numbers under zero, and the density page names both together: this one, which is healthy data, and a channel that hit the scanner's black level, which is not.
The linked and luma sets keep a plain 0 to 1, because they read the per-channel windows' output: a dimensionless signal with no unit to state. The readouts follow (three decimals of density on the three channel windows, a percentage on the two global ones) and every number quoted below says which of the two scales it is on.
The histogram is measured before the handles
This sounds like an implementation note and it decides whether the tool is usable. The histogram you place handles against is measured upstream of the levels. If it were measured after, it would deform as you dragged, and "put black at the edge of the data" would be chasing a moving target. Photoshop does the same, for the same reason.
A consequence worth knowing: moving a levels handle triggers no new measurement at all. Only a change of gain does.
The scan skips the paper, and the midtone puts it back
A colour negative is exposed at a gamma of roughly 0.6: a deliberately flat recording, because nobody was ever meant to look at it. The paper it was designed for is the opposite: RA-4 colour paper works around 2.5 to 3, and the two multiply. The chain lands near 1.6, and that is the contrast a print actually shows you. Scanning skips the paper entirely, so a scan that has only been inverted is missing a step of the process, not badly adjusted.
Putting that step back is what the resting midtone does. On a negative it sits at 0.670 on the linked track, which runs from 0 to 1, and not on the per-channel tracks graduated in density. That position is an exponent of 1.73, which I measured identically across every negative reference frame. It lands where film and paper together land, but the datasheets are not where it came from: it came off the frames, and it would still be 1.73 if the arithmetic above disagreed.
Where 0.670 comes from, if you want the exponent
1.73 is the exponent, and 0.670 is where it puts the middle of the scale: 0.5 raised to the power 1/1.73 gives 0.670. The handle reads the position, not the exponent, which is why the number on screen is 0.670 and not 1.73.
There is nothing of the sort on a positive, which was never printed, so there the resting midtone is the true neutral. The app knows the difference, because it knows which mode the frame is in.
On the linked track, that resting value is drawn at mid-travel while still reading 0.670, because a rest sitting two thirds along its track would read as already pushed, every time you looked at it. So a handle in the middle means the setting is at rest, right of the middle means more contrast, and the number beside it is there to compare two pictures rather than to read off the position.
Only the linked midtone is anchored that way. The per-channel tracks stay linear, because there you are placing points on data, and one curved handle in front of straight landmarks would be two rulers graduated differently on one drawing.
Five points, not three
Each set carries black, shadows, midtone, highlights and white. The two extra points exist because of something a single gamma cannot fix: once the medians of the three channels are aligned exactly, the film's three characteristic curves still leave the tenth and ninetieth quantiles apart, because they do not climb at the same rate. One exponent cannot fold both ends. Two more control points can.
The inner three are stored relative to black and white, so moving an end point carries them with it and the shape you dialled survives. The ends themselves are pinned: a black and a white placed on data keep their meaning whatever the inner three are doing.
Nothing here is applied for you
The resting positions are where every frame opens, on purpose. They are not a preset and not a correction: no film base is detected, nothing is read from your image to decide them. The automatic balances write into these same five handles when you click one, and ⌘Z takes it back.
Density figures and the 0.670 resting midtone measured on the eight reference films I scanned to calibrate the application, 2026. They are named on the automatic balance page, together with what is not in that set.
In the manual: Per-channel levels
Place a black point where the data is
Open a frame and the handles read in real densities immediately, and there is no calibration step and no film to look up. If you have never converted a negative at all, the six-step guide gets you to a finished picture without touching any of this.
Next: the six automatic balances, which write into these same five handles in one click. The glossary holds the words used here.
Free, no account, no trial limit. Real densities from the first frame you open.