Why Color Negatives Turn Blue When Inverted: The Math of C-41 Orange Mask Subtraction

Load a camera scan into any generic photo editor, hit *invert*, and the faces come back cyan. The usual diagnosis — "the white balance is off" — is wrong, and that is why chasing it with the temperature slider never quite lands. The cast is baked into the film on purpose, and removing it is a division, not a subtraction.

1. Why the film base is orange at all

The cyan and magenta dyes in a colour negative are not spectrally pure: each absorbs light in bands it should ignore. Left alone, that unwanted absorption contaminates every colour reproduced from the negative. Kodak's fix, standard across C-41, is colored couplers — the unused coupler stays yellow-orange and forms a positive mask of exactly the contamination it is compensating for. The orange you see is a built-in correction filter, not a defect.

2. The linear trap: 1 − RGB

Film responds to light exponentially. Optical density is D = log₁₀(1 / T) where T is transmittance, and the emulsion's answer to exposure is linear in *density*, not in transmittance. Inverting an 8-bit sRGB value with 255 - x inverts transmittance, so the orange bias survives — inverted into cyan — while the shadows, which occupy the compressed end of the transmittance scale, clip into a flat navy block.

3. Subtracting the mask in density space

In density space, removing the base is a plain subtraction: D_image = D_pixel − D_base. Exponentiate both sides and subtraction becomes division — T_image = T_pixel / T_base — which is why the shader divides by a sampled mask colour instead of subtracting it. One power function then restores the emulsion's non-linear response.

vec3 inverted = pow(max(vec3(0.0), 1.0 - (pixel / mask)), vec3(u_filmCurve));
inverted = clamp((inverted - u_black) / (u_white - u_black), 0.0, 1.0);
inverted = pow(inverted, vec3(1.0 / u_gamma));

• pixel / mask — density-domain subtraction of the orange base.
• u_filmCurve (≈1.45–1.60) — restores the S-shaped emulsion response instead of a flat ramp.
• u_black / u_white — level stretch, so the histogram fills the range without clipping.
• u_gamma (1.8) — compensates the display transfer function; without it midtones sit dark and muddy.

4. Getting `mask` right: 40×40 with outlier rejection

The whole correction rests on one number per channel, so a single dusty pixel would poison the frame. NegaScan averages a 40×40 patch of unexposed base — the sprocket gap or the strip between frames — and rejects samples beyond 1.5 standard deviations before averaging, which removes grain spikes, dust and scratches without hand-masking.

5. Run it yourself

If an inversion tool cannot tell you which pixels it used as the film base, it is guessing — and you will be correcting its guess on every frame.