Post-Production Film Grain — deriving settings from datasheets

Why this file exists. The X-T5's Grain Effect is unavailable in movie mode (video-mode-settings.md). For video, grain has to be added in post, which means the question is no longer "which Grain setting" but "what numbers, in which plugin, to match which stock." Almost nobody answers that from primary data. This file does, for the cinema stocks in X-T5/originals/cinema/.

Scope note: stills-stock grain is set in-camera and is out of scope here by design.

---

The honest summary, first

There are three parts to matching film grain, and they are not equally knowable:

	Can it be derived from datasheets?	Status
Grain scale (how big, in pixels)	✅ Yes — exactly. Pure geometry from a number printed in every datasheet	Solved below
Relative intensity between stocks	✅ Yes — from published RMS granularity, where Kodak prints a number	Solved below
Absolute plugin values (e.g. "Lightness = 14")	❌ No. Neither Adobe nor Blackmagic publishes the units or transfer function of their grain parameters	Calibration procedure given instead

Anything that hands you an absolute number for these plugins without a calibration step made it up. What is derivable is the scale and the ratios, and those are the parts that actually make one stock distinguishable from another.

---

Part 1: The 48-micrometre anchor

Every granularity figure in this bank's archive. Kodak, Ilford, Fujifilm, stills and cine alike. Carries the same footnote:

"Read at a net diffuse visual density of 1.0, using a 48-micrometer aperture." Eastman Double-X 5222 datasheet "Read with a microdensitometer, using a 48-micrometre aperture." Kodak Vision3 250D / 500T / 50D / 200T, Ektachrome 100D, 2383

That is the ISO-standard measurement window for diffuse RMS granularity. It is constant across every stock and every manufacturer, which makes it the one dimension that can be carried from a datasheet into a timeline.

A necessary precision: 48 µm is not the size of a grain clump. Grain structure is smaller; the aperture is the sampling window through which density fluctuation is measured. What it gives us is the spatial scale at which the industry agreed graininess becomes perceptually meaningful, and that is exactly the scale a grain plugin should be operating at.

The projection

Project 48 µm onto the delivery raster and you get the scale, in pixels, that a matched grain should occupy:

Format	Frame width	@1920	@2048 (DCI 2K)	@3840 (UHD)	@4096 (DCI 4K)
Super 35 (4-perf)	24.89 mm	3.7 px	3.9 px	7.4 px	7.9 px
Super 16	~12.35 mm	7.5 px	8.0 px	14.9 px	15.9 px
35 mm stills	36 mm	2.6 px	2.7 px	5.1 px	5.5 px
X-T5 (APS-C, 23.5 mm)	23.5 mm	3.9 px	4.2 px	7.8 px	8.4 px

(X-T5 at its native 7728 px stills raster: 15.8 px.)

Three consequences, and all three are things people get wrong:

1. Grain scale doubles when you double the raster. A grain setting dialled in on a 1920 timeline is wrong by a factor of two at 3840. This is the single most common film-grain mistake in editing, and it is why grain that looked right in the edit reads as digital noise on a 4K export. Set grain at delivery resolution, or scale the size parameter with the raster.

2. Super 16 needs ~2× the grain scale of Super 35 at the same delivery resolution (24.89 / 12.35 = 2.02). This is the quantitative version of something this bank already states qualitatively, "format dominates apparent grain" (validation-methodology.md). 16 mm does not look grainier because of a different emulsion; it is frequently the same emulsion, enlarged twice as much. Vision3 500T in Super 16 and Vision3 500T in Super 35 are the same film and should get the same intensity and double the scale.

3. The X-T5 sits almost exactly on Super 35. 23.5 mm against 24.89 mm, a 5.6 % difference, well under one pixel at UHD. So for video work, X-T5 footage and Super 35 origination want effectively identical grain geometry, which makes the cinema recipes' grain translation unusually clean. That is a genuinely convenient accident of APS-C sizing and worth knowing.

Caveat on the dimensions: Super 35 at 24.89 × 18.66 mm is the widely-used figure, but there was never an official SMPTE standard for Super 35. Camera manufacturers vary slightly. Super 16 varies more (12.35 mm here; sources give 12.35–12.52 mm depending on SMPTE vs ASC vs specific camera). Neither variation exceeds ~1.5 %, which is sub-pixel at these rasters, so the table stands, but the numbers are nominal, not certified.

---

Part 2: Relative intensity, from published granularity

Kodak's colour motion-picture datasheets do not publish a single RMS number. They print a granularity curve (Sigma D against density) with the instruction to "find the density on the left vertical scale… read the number and multiply by 1000 for the rms value." That is a chart, and this bank has not digitised those charts, so no RMS figure is claimed for any colour Vision3 stock, or for Ektachrome 100D, or for 2383. Stating one would be invention.

One cine stock does print a bare number:

Cine stock	Diffuse RMS granularity	Source
Eastman Double-X 5222	14	Datasheet, at net diffuse visual density 1.0, 48 µm aperture
Vision3 50D / 250D / 200T / 500T	curve only	Not digitised — no number claimed
Ektachrome 100D 5294	curve only	Not digitised — no number claimed
Vision 2383 print	curve only	Not digitised — no number claimed

Double-X also publishes resolving power: 32 lines/mm at TOC 1.6:1, 100 lines/mm at TOC 1000:1, the only cine stock here with that figure.

The RMS ladder, for calibration reference

RMS granularity is a linear scale, so ratios are meaningful. Drawing on the stills stocks whose sheets do print numbers (all measured through the same 48 µm aperture, so directly comparable):

Stock	RMS	Relative to Double-X
Fujicolor Pro 400H	4	0.29×
Fujifilm Acros 100	7	0.50×
Kodak Ektachrome E100	8	0.57×
Kodachrome 25	9	0.64×
Kodachrome 64	10	0.71×
Eastman Double-X 5222	14	1.00×
Kodachrome 200	16	1.14×

Use this for the intensity ratio between stocks, not for absolute values. If you calibrate a grain plugin so Double-X reads correctly, Acros should be set to roughly half that intensity, and Kodachrome 200 to about 1.14×.

⚠Do not mix scales. Kodak's stills sheets for the modern colour negatives use Print Grain Index (Ektar <25, Portra 400 = 37, UltraMax 400 = 46, threshold of visibility = 25, four units per just-noticeable difference), and Kodak states explicitly that PGI "cannot be compared to rms granularity." They are different measurements on different scales. Rank within PGI, rank within RMS, never across.

---

Part 3: The tools, verified

DaVinci Resolve: free version

⚠ResolveFX Film Grain is Studio-only. It is not available in the free version. Two routes remain:

A. The Fusion Film Grain node (available in free Resolve, Fusion is not restricted) Add a Fusion composition, then search for Film Grain (Fgr). Its documented controls:

Control	What it does	How to set it from this file
Presets: 8mm / 16mm / 35mm	Preset grain profiles by format	Start from 35mm for Super 35 origination; 16mm for Super 16. This preset is the format-scale decision — it is doing the Part 1 geometry for you, approximately
Red / Green / Blue curves	Per-channel grain amount	Set blue highest, red lowest. Documented behaviour: "each colour component of film presents a different grain profile, typically with the blue channel presenting the most significant amount of grain." This matches colour-negative construction, where the blue-sensitive layer sits on top
Log Processing	Toggles grain response curve	ON when matching film. The docs are explicit: "enabled when matching film, and disabled when working with images that require a more linear grain response"
Offset	Grain intensity in deep blacks	Raise slightly if blacks look too clean. Real film grain does not vanish in shadow the way synthetic noise does
Seed / Reseed	Randomisation	Change per shot if grain patterns visibly repeat across a cut

B. Grain-scan overlays, a scanned real-film grain plate on a track above, set to Overlay or Soft Light blend mode. This is the most authentic option available in free Resolve, because it is a photograph of actual grain rather than a synthesis of it. Scale the plate so its grain structure matches the Part 1 pixel figure for your format and raster.

Premiere Pro

⚠The obvious effect is the wrong one. Video Effects > Noise & Grain > **Noise** has only an Amount of Noise percentage, no size control at all. It produces uniform per-pixel noise, which is not what film grain looks like at any scale.

Use Noise HLS Auto instead (same Noise & Grain category). Its documented parameters:

Parameter	Set to
Noise (type)	Grain — this is the setting that activates the size control. Leave it on Uniform and you have the wrong effect again
Grain Size	Active only when Noise = Grain. This is your Part 1 scale parameter
Lightness	The main intensity control — grain as luminance fluctuation, which is what film grain predominantly is
Saturation	Small. Real grain has some chroma variation, but colour-negative grain is mostly luminance
Hue	Near zero. Hue noise reads as digital chroma artefacting, not film
Noise Animation Speed	Non-zero — static grain over moving footage reads immediately as fake ("fixed-pattern noise"). Real grain is re-randomised every frame

Units are not published. Adobe documents what each parameter does but not its range or transfer function, so any specific number here would be a guess. Calibrate instead, see below.

---

Part 4: Calibration procedure

Ten minutes, once per project, and it converts the ratios above into real numbers for your actual pipeline.

1. Work at delivery resolution. Not the edit proxy. Part 1 shows the scale is raster-dependent.
2. Find your reference. Best: a real scan of the stock (see _reference-sources/). Failing that, the cinema originals' the notes files describe the expected grain character in words.
3. Set the size first, intensity second. Look up your format and raster in the Part 1 table. In Fusion, pick the matching format preset; in Premiere, adjust Grain Size until the grain structure visually occupies about that many pixels. Zoom to 100 %. Grain judged on a fit-to-window viewer is judged at the wrong scale, which is the second most common mistake after point 1.
4. Then dial intensity on a mid-grey. Grain is measured at density 1.0, a mid-tone, so judge it there, not in highlights or shadows.
5. Write down the value you land on. That is your pipeline's anchor. Every other stock can then be set from the RMS ratio in Part 2 without recalibrating.
6. Check per-channel. In Fusion, blue > green > red. Premiere's Noise HLS Auto has no per-channel control, a real limitation, and the strongest argument for using Resolve's Fusion page for grain even if you edit in Premiere.
7. Check motion. Scrub. Static grain is a giveaway.

---

Part 5: Per-stock notes for the cinema originals

Grain character to target	Basis
Heaviest in the cine set. RMS 14 — coarser than Kodachrome 64 (10), close to Kodachrome 200 (16). Resolving power 32 lp/mm at TOC 1.6:1	✅ Datasheet number
Finest. The datasheet's finest-grain claim; the bank's scan analysis found grain "very nearly absent" on a Nikon F100 + 50/1.4 frame	Datasheet claim + scan
Fine, moderate	Datasheet claim only
Most grain of the Vision3 line — a 500-speed negative	Speed class
Fine — reversal stock, and reversal grain reads differently: it is densest in the highlights, the inverse of negative film	Datasheet + reversal physics
Special case — see below	—

The 2383 special case, and why it matters for anyone chasing a real film look

2383 is a print stock, not a camera negative. In a real photochemical finish the grain you see on screen is the sum of camera-negative grain and print-stock grain, with the print stock adding its own on top of an already-grainy image.

Practical consequence: if you are grading toward the 2383 Print recipe, the theatrical projection look, the honest grain structure is two layers, not one: a camera-negative grain matched to whichever stock you are notionally shooting, plus a second, finer print-grain pass. A single grain layer will always read slightly too clean or slightly too coarse, because it is trying to represent two stacked processes with one.

This is the kind of thing a single "film grain" plugin preset cannot express, and it is worth knowing before concluding that a grade is not landing.

---

What is still missing, stated plainly

1. The Vision3 granularity curves are still not digitised, but the attempt has been made and documented. See the appendix: the axis calibration is now solved and verified (183 px/decade at 288 dpi, three charts agreeing within 1 %), and two findings came out of it. B > G > R confirmed from Kodak's own data, and granularity varies with density, so no single RMS number fully describes a colour negative. What remains is manual point-picking with a graph digitiser, plus an explicit decision on the net-vs-absolute reference density. 200T may be unrecoverable from its datasheet, its curves overlap in the source artwork.
2. No plugin units are published by Adobe or Blackmagic, hence the calibration procedure rather than a settings table. If someone measures the transfer function of Noise HLS Auto's Lightness parameter, absolute values become derivable.
3. No grain-matched test footage exists in this bank. The Part 1 geometry is exact, but it has not been visually confirmed against a real film frame at matched resolution.

---

Appendix: the Vision3 curve digitisation attempt (2026-07-26)

The task flagged above as "the highest-value outstanding item in this file" was attempted. It partly succeeded and partly failed, and both halves are recorded, the failure is more useful than a set of confident-looking numbers would have been.

What was done

The four Vision3 datasheet PDFs were rendered at 288 dpi (pypdfium2, scale 4), the granularity chart isolated, and the plot box and axis tick marks located by pixel analysis rather than by eye.

✅ Established: the axis calibration, and it is exact

The GRANULARITY SIGMA D axis is logarithmic and maps linearly onto the DENSITY axis. Measured independently on three charts:

Stock	px per decade	Density units per decade
Vision3 50D	182.5	0.941
Vision3 250D	183.0	0.940
Vision3 500T	184.2	0.940

Three independent charts agreeing to within 1 %. On the 250D chart the ladder was checked across two full decades (0.001 → 0.01 → 0.10) and the spacing was identical to the pixel: 183 px each. So the conversion is:

σD(y) = 0.001 × 10^((y₀ − y) / 183): where y₀ is the pixel row of the 0.001 tick, at 288 dpi.

Anyone finishing this work can use that directly.

✅ Established: B > G > R, confirmed from Kodak's own data

On every chart the granularity curves are ordered blue highest, green middle, red lowest, across essentially the whole exposure range.

This independently corroborates, from a film manufacturer's primary data, the claim taken earlier in this file from Blackmagic's Fusion documentation, that "each colour component of film presents a different grain profile, typically with the blue channel presenting the most significant amount of grain." Two entirely unrelated sources, a 2022 Kodak datasheet and a compositing application's node reference. Agreeing on channel ordering.

So the per-channel guidance in Part 3 is more than software convention: it is a property of the film. Set blue > green > red with confidence.

✅ Established: there is no single "RMS number" for a colour negative

The granularity curves are not flat. On every Vision3 stock granularity rises through the mid-scale, peaks, and then declines at high density. That is why Kodak prints a curve here and a single figure on the black-and-white sheets: for a colour negative, granularity is a function of density, and any single number is a one-point summary of that function.

Consequence for grain matching: a correctly matched grain should in principle be density-dependent. Stronger in the mid-tones than in the deep shadows or the highlights. The Fusion Film Grain node's Offset control (which adjusts grain intensity in the deep blacks) exists for exactly this reason, and this is the primary-source justification for using it.

❌ Failed: per-stock RMS values

No RMS figure is published here, because the extraction was not reliable enough to publish. Two independent reasons:

1. Automated curve tracing broke on the dashed lines. The granularity curves are dashed on most of these charts, so a continuity tracker loses them at every gap and swaps between curves where characteristic and granularity curves cross. Runs returned 3–6 curves where 6 were expected, and produced at least one obviously impossible value (an "rms 952"). A tracer that fails visibly is fine; one that fails quietly would have poisoned the ladder.

2. The four granularity charts are not drawn to a common template. This was the genuinely surprising find, and it invalidates any single extraction script for them. ⚠Scope correction (later on 2026-07-26): this applies to the granularity charts only. It is NOT a property of the datasheets as a whole. The sensitometric charts in the very same four PDFs share an identical axis (Camera Stops −8…+8) and traced cleanly on all four stocks at 100 % ordering validation. Judge the chart, not the datasheet.

	50D · 250D · 500T	200T
Granularity curves	dashed	bold solid
Density axis labels	every 1.0	every 0.2
Sigma D ticks detected	15	11
Channel separation	clearly separated	B, G and R overlap into one band

On the 200T chart the three granularity curves are drawn thick and lie almost on top of one another through most of the exposure range. They are not visually separable at any rendering resolution, because the information is lost in the source artwork, not in the rendering. That is a limit of the document, not of the method.

What would actually finish this

• A dedicated graph digitiser (WebPlotDigitizer or equivalent) with manual point picking, using the verified 183 px/decade calibration above. Manual picking handles dashed lines and crossings that automated tracing cannot.
• A reference-density decision that has to be made explicitly. Kodak's stills sheets specify "read at a net diffuse visual density of 1.0" Net, i.e. above D-min. The cine sheets say only "find the density on the left vertical scale." For a colour negative with an orange mask, the three channels have very different D-min values, so "density 1.0" means something different per channel and the reading changes materially depending on which convention is used. This ambiguity must be resolved and stated before any number is published, or the resulting ladder will not be comparable with the stills-sheet RMS figures in Part 2.
• 200T may be unrecoverable from the datasheet and might need a different Kodak publication (H-845, The Essential Reference Guide for Filmmakers, is referenced by all four sheets and may tabulate what the charts only plot).

Until then Part 2 is written: Double-X 5222 (rms 14) remains the only cine stock in this bank with a published granularity figure, and no value is claimed for any colour Vision3 stock.

Postscript: the same technique then worked, on a different chart (2026-07-26)

The tracer that failed on the Vision3 granularity curves was pointed at the Kodak 2383 print stock's sensitometric chart, and succeeded cleanly. Full account in 2383 validation.md.

That contrast is the actual methodological finding here, and it is worth stating plainly: the technique was never the problem, the source charts were.

	Vision3 granularity curves	2383 sensitometric curves
Line style	dashed — breaks continuity tracking	solid
Curve separation	overlapping, 200T bold and inseparable	cleanly separated through the rising band
Template consistency	four sheets, no common template	single well-drawn chart
Tick calibration	solved (183 px/decade)	solved (128.0 px/log E, 128.33 px/density)
Ordering validation	not achievable	✅ 93/93 columns correct
Independent cross-check	none available	✅ system gamma 1.89–2.07 vs the expected ~1.8–2.0
Outcome	❌ no numbers published	✅ γ 3.44, D-max 4.09, 4.4-stop scale

The lesson for future digitisation work in this bank: before attempting a trace, check whether the chart is solid-line, separated, and independently cross-checkable. If it is, the numbers can be trusted to ~1 %. If it is not. Dashed, overlapping, or drawn to an inconsistent template, the honest output is a documented failure, not a plausible-looking number. Both outcomes now exist in this repo as worked examples.

Second postscript: the tracer is now calibrated against published ground truth (2026-07-26)

The 2383 result rested on an indirect cross-check (system gamma landing in the expected range). Double-X 5222 provided a direct one, five times over.

Its datasheet plots five curves and prints the gamma of each, 1.05, 0.84, 0.66, 0.56, 0.50, because the chart exists to show how development time changes contrast. Traced blind to those labels and then compared (full account):

Mean absolute error 0.016 · maximum 0.018 · correlation r = 0.9980 (n = 5): published gamma reproduced to about 2 %, five times over.

(Correlation corrected 2026-07-27 from a published 0.99908 that does not reproduce; r is also the wrong statistic for this test, MAE carries the claim. See the notes §6.)

This turns the tracer from "worked once" into a calibrated instrument, and it retroactively supports every figure in the 2383 measurement.

It also calibrated the method's dominant error term, the fit window. Gamma is the slope of the "straight-line" portion, but that portion has no exact boundary, so the answer depends on the window width. Tested against the five known values, the error is U-shaped with two distinct, directional failure modes:

• Too narrow (≤0.5 log E) biases high: taking a maximum over many noisy slope estimates is upward-biased by construction.
• Too wide (≥1.2 log E) biases low: the window outruns the straight section and drags in toe and shoulder. At 1.5 log E the γ=1.05 curve collapses to 0.80.

0.8 log E is the optimum (mean error 0.016) and is now this bank's standard fit window: calibrated against published values rather than chosen by eye.

Anyone re-deriving a gamma from a datasheet in this repo should use 0.8 log E and say so, because quoting a gamma without its fit window is quoting a number without its units.

---

Sources: archived Kodak motion-picture datasheets in X-T5/_reference-sources/datasheets/kodak-motion/ (read directly, 2026-07-25); Blackmagic and Adobe feature documentation and Fusion node references via web search; Super 35 and Super 16 frame dimensions cross-checked and reported with their known variation. Software availability (ResolveFX Film Grain = Studio-only; Fusion Film Grain available in free Resolve) verified 2026-07-25.

Last updated: 2026-07-26