dtourolleandClaude Opus 5
4b2ee0ac50
An emulsion is a suspension of crystals. Light sensitises some; development
turns a sensitised one opaque, all or nothing. So a patch of film's density
is a *count* of developed grains, and a count of independent yes/no events
has a variance whether or not anyone wanted texture:
mean = D
variance = D * (Dmax - u * D) / N
That expression is the whole feature. It peaks in the middle of the density
range and vanishes at both ends -- clear film has nothing developed to vary,
black film has nothing left to develop -- so grain lives in the midtones as a
consequence rather than as a "midtone bias" slider.
I was wrong earlier that this needs the detail stage. Nothing in it reads a
neighbouring pixel; the only reason to move it was that grain must be fixed in
film space rather than screen space, and that solves itself: N is grains *per
pixel*, so it scales with the film a pixel covers. Zoom out, each pixel
averages more grains, less variance -- correct, with nothing super-sampled and
nothing filtered. It stays in the fused pass.
Grain goes on the density and *before* the dye, which is the physical order
and not cosmetic. Perturbing the finished colour -- what an effect does --
tints highlights wrong, because that noise never passes through the dye.
Crystal habit lives in `rms_granularity`, the number every datasheet
publishes, now a profile field. It measures exactly what differs between a
cubic emulsion and a tabular one: at equal speed, tabular crystals present
more area per unit silver, so the film reads finer. Delta 100 is quoted near 9
where HP5 is near 12, and that gap *is* the habit. Adding a stock whose grain
is its whole reputation is therefore editing one line, not writing a model.
Three things this cost, all of them worth writing down:
- The default granularity is a colour negative's, blue coarsest. Applied to
Tri-X it put *colour* speckle on a black and white photograph. Monochrome
stocks collapse it at parse, where every other per-layer table is already
replicated from the one measured channel.
- Helpers cannot read uniforms. The composer prefixes a uniform with its
operation's id and rewrites references inside a fragment body only;
helpers are shared and deduplicated, so a bare `gn0` names nothing.
`film_lut` already took its size as an argument for this reason, and now
says so.
- The end-to-end test compares the shader against the CPU model, and grain
is stochastic, so that comparison now runs with grain off. Which means a
grain that never left the CPU would look exactly like a passing suite --
hence a second test that grain off is bit-identical, one grain per pixel
moves it, and ten thousand move it less.
Not here, deliberately: no grain slider. The parameters are physical and
`rms_granularity` is the honest place to scale one from, but its range wants
choosing rather than guessing. Nor a film format -- 35 mm is assumed, and
medium format at the same stock is far less grainy per unit of picture.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Film stocks
One file per stock in profiles/. Adding a stock is adding a
file — no code change, no shader, no new operation — for the same reason
dr-decode's base curves work that way: under the GPLv3 a stock should be
contributable without a release.
What a profile is
Three measured tables, all of them published in the manufacturer's datasheet:
| Field | What it decides |
|---|---|
log_sensitivity |
what each emulsion layer sees, per wavelength |
density_curves |
contrast, latitude, and where the stock clips |
dye_density |
what the developed stock looks like, per wavelength |
base_density |
the support: film base, and a colour negative's orange mask |
Plus kind (negative or positive), support (film or paper), and the two
illuminants the data is referenced to. A print paper is a stock like any
other; support exists so an interface can offer papers separately, not
because the renderer treats them differently.
Why it is not a LUT
Because the parameters stay physical. Opening up a stop moves the picture along the film's own characteristic curve — toe, shoulder and all — instead of scaling a number somebody baked at one exposure. A scanned negative comes out orange and inverted because that is what a negative is, and it becomes a photograph when a paper profile prints it, exactly as it would in a darkroom.
The data cost runs the other way from a LUT collection too: a stock is about 17 kB of measurements, where one HaldCLUT is roughly 800 kB of one person's grade.
How it runs
The spectral chain reduces to three tables, and the reduction is exact where it
matters — see src/bake.rs for the argument:
- A 3×3 matrix, linear sRGB to the three layers' exposure. Exact, not an approximation: the reconstructed scene spectrum is linear in the sRGB triple, so the integral collapses into nine numbers.
- Three 1D curves, log exposure to density, sampled at 256 points.
- One 32³ lookup, density to linear sRGB — dye absorption, the print through the negative, the paper, the viewing illuminant and the chromatic adaptation, all of which take exactly three numbers in.
Per pixel that is a matrix multiply, three curve taps and one texture fetch. Splitting 2 from 3, rather than baking one LUT over exposure, is measured rather than assumed: the curve carries all the sharp shape and the dye mixing is smooth, so folding the curve into the 3D lookup would need it three times larger for the same error. At 32³ the worst interpolation error is about 0.003 in linear sRGB, below one 8-bit code value, and there is a test that says so.
Adding a stock
If spektrafilm has it, add its name to STOCKS in
tools/film-profiles/convert.py and
re-run it. Otherwise write the YAML by hand from the datasheet; the loader
validates the table lengths and says which file and field is wrong.
Either way, list it in BUILT_IN in src/lib.rs to compile it
in — or drop it in the profile directory at runtime, which is the path meant
for stocks that ship separately from the binary.
Provenance
The shipped profiles are converted from
spektrafilm by Andrea Volpato,
licensed CC BY-SA 4.0. See profiles/LICENSE-PROFILES.txt
for the licence and profiles/CHANGELOG.txt for what
the conversion changed and what it deliberately did not.
The sRGB reflectance basis is Mallett & Yuksel (2019); the observer is the CIE 1931 2°.