All posts
By Pravit Gandhi··11 min read

How to grade F-Log and F-Log2 footage

How do you grade Fujifilm F-Log footage? Start from Fujifilm's published code values, work out which curve you shot, then normalize before you look.

Normalize with a transform built for the exact curve you shot, correct exposure and white balance, then build the look. The only Fujifilm-specific complication is that "F-Log" now names three things, and they do not share a curve. Fujifilm publishes reference values for all of them, so you can check rather than guess: on a 10-bit full-range scale, F-Log puts a black patch at code 95, an 18 percent grey card at 470 and a 90 percent white card at 705. F-Log2 and F-Log2 C put the same three at 95, 400 and 570.

What Fujifilm publishes

Unusually for a camera maker outside the cinema tier, Fujifilm publishes complete data sheets: anchor code values, IRE equivalents, the encoding formula with its constants, and the inverse. The F-Log sheet describes the design intent as following "the density of negative films, which has a high compatibility with post production technique fostered in the field of cinema film."

The published anchors, on Fujifilm's own 10-bit scale with the accompanying IRE column:

Input reflectionF-Log IREF-Log codeF-Log2 IREF-Log2 code
0 percent3.5953.595
18 percent4647038400
90 percent7370558570

Two details there repay reading carefully. The code values are full range, 0 to 1023, while the IRE column is legal-range referenced, so the two columns are not the same measurement written twice. That is the convention Sony and Panasonic use and not the one in ARRI's LogC4 specification, which is a reliable way to make cross-manufacturer comparisons go silently wrong.

The second is that F-Log and F-Log2 disagree about middle grey by 70 code values. If your normalization assumes the wrong one, the error is not a flat offset: it varies across the range and is worst in the highlights. F-Log2 footage decoded as F-Log is 0.182 stops out at 2 percent reflectance, 0.740 at middle grey and 1.315 at a 90 percent white card. The reason is what F-Log2 was built to do — it fits three more stops of highlight into the same container, so the two curves have to separate further and further as you go up, while below grey they nearly coincide. Their toe cuts sit within two thousandths of a stop of each other.

We took Fujifilm's formulas through the same validation used for our nine-format code value study, which now includes both Fujifilm curves, implementing each verbatim and confirming it reproduces the manufacturer's own anchors first. All six Fujifilm anchors reproduce within half a code value, so the figures below are ours, computed from Fujifilm's formulas rather than from Fujifilm's marketing.

Where F-Log2 differs from F-Log

Fujifilm's manual says only that F-Log2 and F-Log2 C are "gamma curve options with a wider dynamic range than [F-Log]." The computation says how much wider, and what it costs.

In its straight-line region F-Log spends about 106 code values per stop. F-Log2 spends about 76. That is the whole trade in one pair of numbers: F-Log describes each stop more finely, F-Log2 fits more stops into the same container. Following each curve up until it reaches code 1023, F-Log gets about 5.3 stops above middle grey and F-Log2 about 8.3, so three extra stops of encoded highlight range bought by describing every stop around 30 percent more coarsely.

That places the two in different company. F-Log's slope sits close to Nikon N-Log at about 104 and Canon Log 3 at about 113, both steep through the mid-range and comparatively short at the top. F-Log2 sits with the mainstream group: Panasonic V-Log at about 74, ARRI LogC3 at EI 800 at about 76, Sony S-Log3 at about 79. If you have graded S-Log3 or V-Log, F-Log2 will feel familiar in a way F-Log will not. See how to grade N-Log footage and how to grade V-Log footage.

One caveat, because the figure is easy to misuse. Where a curve reaches code 1023 is a property of the encoding maths, not a statement about where any Fujifilm camera clips, and sensor clipping is a separate body-specific number Fujifilm does not publish.

F-Log2 C is the third option and differs by gamut rather than by curve. Fujifilm's data sheet states that "the gamma curve of F-Log2 is identical to that of F-Log2C," and the GFX ETERNA guide confirms it by printing the same formula constants for both. What changes is the primaries: F-Log and F-Log2 use F-Gamut, whose primaries match ITU-R BT.2020, while F-Log2 C uses the wider F-Gamut C. A tonal transform built for F-Log2 is therefore right for F-Log2 C, and a gamut transform is not.

Which curve you actually have

Three ways, cheapest first.

Read the metadata. The GFX ETERNA guide's metadata table documents separate labels for F-Gamut and F-Gamut C as colour primaries, and separate transfer characteristic labels for F-Log and F-Log2, with F-Log2 C sharing the F-Log2 label. If your application surfaces this, it is definitive.

Read the scope. Point a waveform at a grey card in an untouched clip. Middle grey at code 470 is F-Log, at 400 is F-Log2 or F-Log2 C.

Read the ISO floor. Fujifilm's X-H2 manual states sensitivity is "restricted to values between ISO 500 and ISO 12800 ([F-Log]) or between ISO 1000 and ISO 12800 ([F-Log2] and [F-Log2 C])," so a clip shot below ISO 1000 cannot be F-Log2.

Fujifilm's own conversion LUTs and ACES transforms

Fujifilm publishes free .cube files from its support site, organised by camera model. The page carries one warning that matters: "Availability of F-Log2 C / F-Log2 varies by model," so download the pack for your body rather than assuming one covers the range.

These are not plain technical conversions. The current sets convert into Fujifilm's Film Simulations, ten of them, which the GFX ETERNA guide describes as variations of a look defined by "the combination of complex factors, such as contrast and color tones." That is a creative decision arriving with your conversion, and a problem if you expected a neutral starting point. The general version of that distinction is in why CST, ACES and LUTs look different.

Fujifilm also publishes ACES Input Device Transforms for its log formats on its technical data page, the better route for mixed-camera work since an IDT gets you into a defined working space rather than straight to a look. Either way, Fujifilm states the monitoring conditions it assumes when its 3D-LUT is applied: colour space Rec.709, colour temperature D65, gamma 2.2.

Bit depth and data level

Fujifilm's manual does not state a bit-depth requirement for F-Log2, and I am not going to invent one. What it states is that the codec sets the depth: H.264 records "at a depth of 8 bits" with 4:2:0 subsampling, and H.265 records "at a depth of 10 bits" with either 4:2:0 or 4:2:2. Recording log into H.264 gives you 8-bit log, the combination that leaves least room, and what that costs is in why banding appears after grading and export.

The other setting to check before trusting any code value is the data level. Fujifilm gives two options. Video range limits 8-bit movies to the span 16 through 235 and 10-bit movies to 64 through 940. Full range gives the whole 0 through 255 and 0 through 1023. The published anchors above are full-range figures, so on a camera set to video range an untouched clip will not read 470 or 400 at middle grey, and that is a range convention rather than a wrong curve.

A working order

Identify the curve first, because everything downstream depends on it and the failure is quiet rather than loud.

Normalize second, with a transform built for that exact curve, and only once. Two transforms stacked gives plastic highlights and crushed shadows, a more common fault than no transform at all because it looks almost right.

Correct exposure and white balance third, as a separate named step. Where it goes depends on your route: with a conversion LUT, correct upstream while the clip is still log, because once values pass through the table the skew is baked in. With a colour space transform or an ACES input transform, correct in the working space afterwards.

Build the look fourth, keeping the moves few and large. Deal with noise here too, after normalization rather than before, since judging noise on a flat picture leads to under-applying, as covered in why log footage gets noisy when you grade it.

Leumos AI, our browser-based grading studio, handles the first two steps identically across every shot in an upload, using an ACES-based input transform for the format, and returns a rendered 4K ProRes or H.265 file. It is in closed beta with a waitlist at leumos.ai.

Frequently asked questions

What are the F-Log and F-Log2 reference code values?

Fujifilm publishes them in its data sheets. On a 10-bit full-range scale, F-Log records a 0 percent black patch at code 95, an 18 percent grey card at 470 and a 90 percent white card at 705. F-Log2 and F-Log2 C record the same three at 95, 400 and 570. The accompanying IRE figures are legal-range referenced, so they are a different measurement rather than the same one restated.

What is the difference between F-Log and F-Log2?

F-Log2 encodes a wider range and describes each stop more coarsely to do it. Computed from Fujifilm's published formulas, F-Log spends about 106 code values per stop against about 76 for F-Log2, and reaches full scale about 5.3 stops above middle grey against about 8.3.

Is F-Log2 C a different curve from F-Log2?

No. Fujifilm's data sheet states that the gamma curve of F-Log2 is identical to that of F-Log2 C, and both are published with the same formula constants. The difference is colour gamut: F-Log and F-Log2 use F-Gamut, whose primaries match ITU-R BT.2020, while F-Log2 C uses the wider F-Gamut C.

Does F-Log require 10-bit recording?

Fujifilm's X-H2 manual does not state a bit-depth requirement for any of the log options. What it states is that the codec determines depth: H.264 records 8-bit 4:2:0 and H.265 records 10-bit. Recording log into H.264 gives you 8-bit log, which leaves the least room to grade.

Does F-Log2 have more shadow detail than F-Log?

No — the opposite, on the measure arithmetic can actually reach. F-Log allocates 375 code values between black and middle grey. F-Log2 allocates 305. Both curves pin black at code 95, so F-Log2 has 70 fewer.

The confident published answer says F-Log2 has more, and it cites a correct premise on the way to the wrong conclusion: middle grey moves from 46.3 to 38.4 on the IRE scale Fujifilm prints beside its code values. That premise is right — Fujifilm publishes 46 and 38, and our implementation of Fujifilm's formulas returns 46.3 and 38.4 — and it is the datum that refutes the claim it is offered in support of. The container is fixed at 0 to 1023 and black is pinned at 95 on both curves. Move grey down from 470 to 400 to make room above it and there are necessarily fewer values left beneath it. The highlight room came from somewhere, and this is where.

Underneath that sits a second confusion worth separating out. "More stops of dynamic range" and "more code values per region" are opposite quantities when the container is fixed. F-Log2 encodes more scene range precisely by describing every part of it more coarsely. It cannot do both, and neither can anything else at 10 bits.

The sharper version of the finding is that F-Log2 gives up shadow precision without giving up shadow reach. Both curves hand over from their logarithmic branch to a linear toe at almost exactly the same place: 7.660 stops below middle grey for F-Log, 7.662 for F-Log2. They describe the same depth of shadow. F-Log2 buys 3.000 stops of highlight room — full scale at 8.338 stops above grey against F-Log's 5.338 — and pays for it entirely out of how finely the shadows are described rather than out of how far down the description goes.

One limit, and it is the reason this answer is narrower than the question. What we count is code values allocated below middle grey. That is a property of the published curve and it is exact. It is not a measure of how much shadow detail survives, which depends on the sensor's noise floor, and we do not model the sensor. If your shadows are noisy at ISO 6400, the curve did not put the noise there and no allocation figure will tell you which body holds up better. These numbers support claims about allocation and precision. They do not support a claim about visible detail.

What happens if I apply an F-Log LUT to F-Log2 footage?

It reads 0.740 stops dark at middle grey. Reverse the mistake — F-Log footage read as F-Log2 — and it reads 0.969 stops bright. Same 70-code-value gap between the two curves, and it does not cost the same in both directions. The asymmetry is the useful part.

The gap is fixed, but the error belongs to whichever curve is doing the reading, not to the gap. F-Log2 footage puts middle grey at code 400. F-Log's inverse reads that 400 as sitting 70 code values below its own grey of 470, and F-Log spends about 93 code values on the stop below grey, so 70 of them is roughly three quarters of a stop. Going the other way, F-Log's grey arrives at code 470 and F-Log2's inverse reads it as 70 above its own grey of 400 — but F-Log2 spends only about 72 code values on the stop above grey, so the same 70 values buy nearly a full stop. The coarser curve charges more for the same mistake.

This is a mistake Fujifilm shooters actually make, rather than a hypothetical one, because the camera menu offers both curves and a clip that arrives without its metadata does not announce which you chose. It is also sized to survive review. Three quarters of a stop is wrong enough to matter and small enough to be absorbed by an exposure correction that makes the shot look plausible, at which point the grade gets built on top of it. The cheap check is the one in which curve you actually have: put a waveform on a grey card in an untouched clip and read 470 or 400.

Sources