Reference

Log code values, format by format.

Pick a format and a scene target. This page returns the 10-bit code value, the full-range percentage and the legal-range IRE — the three numbers that get mistaken for each other — computed live from each manufacturer's published encoding formula.

Log format
Scene target
-8+8

Sony S-Log3 at 18% grey card — 0.18 reflectance, 0.00 stops from middle grey.

10-bit code value
420
round(encoded × 1023)
Full range, 0–1023. Unrounded 420.00.
Full-range %
41.06%
CV ÷ 1023 × 100
The encoded signal as a share of the full 10-bit scale. ARRI's LogC4 appendix prints this as its IRE column.
Legal-range IRE
40.64%
(CV − 64) ÷ 876 × 100
The convention Sony and Panasonic print beside code values in their own tables.

All three describe the same signal. They are not interchangeable numbers: code value 95 is 9.29 full-range percent and 3.5 legal-range IRE. Copying an IRE figure from one manufacturer's table into another's is the most common way this goes wrong.

Stop ladder — Sony S-Log3

Twelve stops of scene light around an 18% grey card, on the common 10-bit full-range scale.

Stops from greyReflectanceCode valueFull-range %Legal-range IRE
-60.0028111411.1%5.7%
-50.0056213313.0%7.9%
-40.011217116.7%12.2%
-30.022521921.5%17.7%
-20.04527927.3%24.6%
-10.0934733.9%32.3%
00.1842041.1%40.6%
+10.3649648.5%49.3%
+20.7257356.0%58.1%
+31.4465163.6%67.0%
+42.8872971.3%75.9%
+55.7680878.9%84.9%
+611.588686.6%93.9%

All 9 formats at this target

Every format evaluated at 18% grey card, so the only variable is the shape of the published curve.

FormatCode valueFull-range %Legal-range IREPosition on the 0–1023 scale
Sony S-Log342041.1%40.6%
Panasonic V-Log43342.3%42.1%
Canon Log 335134.3%32.8%
ARRI LogC3 (EI 800)40039.1%38.4%
ARRI LogC428527.8%25.2%
RED Log3G1034133.3%31.6%
Nikon N-Log37236.4%35.2%
Fujifilm F-Log47045.9%46.3%
Fujifilm F-Log240039.1%38.4%

The table as data

Every figure here is also published as a file: leumos.ai/tools/log-reference.json and leumos.ai/tools/log-reference.csv. Both are generated at build time from the same code that computes what you see on this page, so they carry the same numbers. They are free to reuse, including commercially, under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/) with attribution to leumos.ai.

Cite it as

Leumos AI (2026). Log format code value reference. https://leumos.ai/tools/log-reference

Mis-tag check

What the wrong tag actually costs.

A mis-tag is not a grade gone wrong. It is footage recorded with one manufacturer's log curve and decoded with a different manufacturer's inverse: the file is intact, the label on it is not. Pick what the footage is and what the editor treated it as, and this returns what that one mistake costs, before anybody touches a wheel.

Footage actually is
Editor treated it as

Footage that is ARRI LogC4, decoded with Sony S-Log3's published inverse.

Error at middle grey
-1.91 stops
log₂(decoded ÷ true scene value)
An 18% grey card is read as 0.0478 reflectance instead of 0.18.
In code values
-135
CV in file − CV the treated-as curve gives grey
The file carries code 285 at middle grey; S-Log3 puts middle grey at code 420.
Reads as
Darker
decoded vs truth
The picture reads 1.91 stops darker than the scene actually was, before anybody grades it.

Across −4 to +4 stops the error runs from 1.27 to 2.47 stops, a spread of 1.20. That spread is the point: it is a different amount of error at every brightness, which is why no single exposure correction fixes it.

Error across the range — LogC4 read as S-Log3

The same wrong decode at nine brightness levels. If a mis-tag were a flat offset, this column would hold one number.

Stops from greyTrue reflectanceCode in fileDecoded asError (stops)Δ code valuesDarker ← → brighter
-40.01121270.00467-1.27-45
-30.02251500.00818-1.46-69
-20.0451850.014-1.68-94
-10.092310.0259-1.80-116
00.182850.0478-1.91-135
+10.363440.0875-2.04-151
+20.724070.159-2.18-166
+31.444710.289-2.32-179
+42.885370.521-2.47-192

All 72 pairings, ranked by size

Magnitude of the middle-grey error in stops, for every ordered pair. Rows are what the footage is, columns are what it was treated as. Reading A as B and reading B as A are different mistakes and do not cost the same, so the matrix is not symmetric. Select any cell to load that pairing above.

Is ↓ / treated as →S-Log3V-LogCanon Log 3LogC3 EI800LogC4Log3G10N-LogF-LogF-Log2
S-Log30
V-Log0
Canon Log 30
LogC3 EI8000
LogC40
Log3G100
N-Log0
F-Log0
F-Log20

Worst pairings

  1. F-Log read as LogC4 — 2.98 stops brighter
  2. V-Log read as LogC4 — 2.41 stops brighter
  3. LogC4 read as V-Log — 2.22 stops darker
  4. S-Log3 read as LogC4 — 2.20 stops brighter

Mildest pairings

  1. F-Log2 read as LogC3 EI800 — 0.00 stops exact
  2. LogC3 EI800 read as F-Log2 — 0.00 stops exact
  3. Log3G10 read as Canon Log 3 — 0.12 stops darker
  4. Canon Log 3 read as Log3G10 — 0.16 stops brighter

The median across all 72 ordered pairs is 0.830 stops at middle grey. The worst is 2.98 (F-Log read as LogC4) and the mildest 0.000 (F-Log2 read as LogC3 EI800, which agree at middle grey and nowhere else). For scale, the largest residual a correct tag leaves behind across these 9 formats at 10-bit is 0.0416 stops.

Notable pairings, at an 18% grey card

Footage isTreated asError at greyVersus quantisation
Fujifilm F-LogARRI LogC42.977 stops72x
Panasonic V-LogARRI LogC42.407 stops58x
ARRI LogC4Panasonic V-Log2.219 stops53x
Fujifilm F-Log2ARRI LogC3 (EI 800)0.000 stops0x

Median across all 72 ordered pairs: 0.830 stops, about 20 times the largest residual that correct normalisation leaves behind (0.0416 stops at 10-bit). The last row is the mildest pairing in the matrix, and it is a special case worth knowing: F-Log2 and LogC3 at EI 800 both put an 18% grey card on code 400, so that one mix-up costs nothing measurable on a grey card — and still costs 0.28 to 0.34 stops four stops down, where the two toes part company. Agreement at one brightness is not agreement.

Why it is never a flat offset

Every one of these curves is a different shape. A mis-tag drops the file's code values onto somebody else's curve, and because the two curves diverge at different rates the recovered scene value is wrong by a different amount at every brightness. That is what the range table shows: the error is largest in the shadows, where the toes disagree most, and changes fastest there. This is why a corrective exposure move, or a curve pulled by eye, closes the gap at one brightness and opens it at another, and why the picture stays stubbornly wrong while looking almost graded.

This is an encoding error and nothing else. It says nothing about how two cameras see: no sensor is being compared here, and a correct tag makes the whole residual vanish to within quantisation. The same footage, correctly interpreted, is the same scene. If the symptom appeared when a LUT went on, the ordering is covered in applying a LUT to log footage in Premiere; for what genuinely separates cameras once the tagging is right, see how far apart cameras really are.

Range conventions and sources

The formats do not all state the same thing about range, and one of them does not measure its input axis the way the others do. Where a manufacturer states a convention, it is quoted; where none is published, that is said plainly.

Sony S-Log3

Range
Full range, stated. Sony: S-Log3 is recorded as full range in XAVC, MPEG and HDCAM SR, and SDI output is full range every time. Sony's own tables print a legal-range IRE column beside the full-range code value.
Scene axis
Reflection, where 0.18 is an 18% grey card.
Source
Technical Summary for S-Gamut3.Cine/S-Log3 and S-Gamut3/S-Log3, appendix “S-Log3 Formula”.

Panasonic V-Log

Range
Full range, stated. Section 3.3 defines in = IN10BIT/1023, so code value = out × 1023. Panasonic's Fig. 2.2 prints a legal-range IRE column beside it.
Scene axis
Input reflection, where 0.18 is an 18% grey card.
Source
V-Log/V-Gamut Reference Manual Rev.1.0, section 3.1.

Canon Log 3

Range
Full range, stated. Canon FAQ Q6: clips are recorded on the card using full range levels, and SDI output is full range.
Scene axis
Scene Linear %, normalised so 100% is a 90% white card — not reflectance. An 18% grey card sits at x = 0.20, not 0.18. Reading the axis as reflectance gives code 339 for middle grey; the correct normalisation gives 351.
Source
Canon white paper “Canon Log Gamma Curves”, 1 Nov 2018, appendix [3][a].

ARRI LogC3 (EI 800)

Range
Full range, stated: 18% grey maps to Log C 0.391, “which is 400/1023”. Exposure-index dependent — ARRI publishes eleven parameter sets from EI 160 to EI 1600; EI 800 is ARRI's default and is what is used here.
Scene axis
Scene linear, where 0.18 is an 18% grey card.
Source
ALEXA Log C Curve — Usage in VFX, 9 Mar 2017, appendix table 2.

ARRI LogC4

Range
Full range, with explicit legal and full code value columns in the specification. ARRI's appendix C “IRE” column is the full-range normalised signal (CV / 1023), not the legal-range IRE that Sony and Panasonic print.
Scene axis
Scene linear, where 0.18 is an 18% grey card.
Source
ARRI LogC4 Logarithmic Color Space Specification, 23 Jan 2025, section 4.1.1 and appendix A.

RED Log3G10

Range
Not stated by RED. RED defines Log3G10 as a normalised float and publishes no integer code value convention. The ×1023 full-range mapping shown here is applied for comparability, consistent with what Sony, Panasonic, ARRI and Canon state for their own curves.
Scene axis
Linear light float, where 18% mid grey is represented as 0.18 — RED's scene axis is reflectance directly.
Source
White Paper on REDWideGamutRGB and Log3G10, form 915-0187 Rev C, “EQUATIONS”.

Nikon N-Log

Range
Not labelled full or legal by Nikon. The specification emits code values on a 0-1023 scale with zero reflectance landing at 127.2, treated here as the full-range 10-bit scale.
Scene axis
Scene linear, where 0.18 is an 18% grey card.
Source
N-Log Specification Document v1.0.0, 1 Sep 2018, section 2.

Fujifilm F-Log

Range
Full range, pinned arithmetically rather than stated in words. The linear branch at zero reflection gives out = f = 0.092864, and Fujifilm's own table prints code 95 for 0% black: 0.092864 × 1023 = 95.00, where the legal-range mapping would give 145. The IRE column printed beside it is legal-range referenced, like Sony's and Panasonic's.
Scene axis
Input reflection, where 0.18 is an 18% grey card — Fujifilm's anchor table is labelled “Input reflection” with rows at 0, 18 and 90, so the axis needs no renormalisation.
Source
FUJIFILM F-Log Data Sheet Ver.1.1, section 2-3.

Fujifilm F-Log2

Range
Full range, on the same arithmetic as F-Log: black at out = 0.092864 × 1023 = 95, matching Fujifilm's printed code 95. The accompanying IRE column is legal-range referenced.
Scene axis
Input reflection, where 0.18 is an 18% grey card. The data sheet states the F-Log2 gamma curve “is identical to that of F-Log2C”, so this curve covers F-Log2 C as well — those two differ by gamut, not by curve.
Source
FUJIFILM F-Log2 Data Sheet Ver.1.1, section 2-3.

What these numbers are

A log curve is a formula. Each manufacturer publishes one: an equation that takes a scene-referred linear value — reflectance, where 0.18 is an 18% grey card — and returns an encoded signal. This page evaluates those published equations directly. Nothing here is measured off footage, fitted to a LUT, or reverse engineered.

Three numbers describe the same encoded signal, and conflating them is the most common error in this subject. The 10-bit code value is the encoded value on a full-range 0–1023 scale. The full-range percentage is that code value divided by 1023. The legal-range IRE is (code − 64) ÷ 876, which is the convention Sony and Panasonic print beside code values in their own tables. Code value 95 is 3.5 legal-range IRE and 9.29 full-range percent. Both are correct in their own document; neither converts into the other by accident. ARRI's LogC4 specification prints an IRE column that is full-range normalised, so an IRE figure copied from one manufacturer's table into another's comparison is wrong before it starts.

The formulas are transcribed from Sony's Technical Summary for S-Gamut3.Cine/S-Log3, Panasonic's V-Log/V-Gamut Reference Manual Rev.1.0, Canon's Canon Log Gamma Curves white paper of 1 November 2018, ARRI's ALEXA Log C Curve — Usage in VFX (LogC3, EI 800) and ARRI LogC4 Specification of 23 January 2025, RED's white paper on REDWideGamutRGB and Log3G10, Nikon's N-Log Specification Document v1.0.0, and Fujifilm's F-Log and F-Log2 Data Sheets Ver.1.1. Each implementation was checked against the anchor values its own manufacturer publishes before any figure here was computed.

Fujifilm is the most completely documented of the nine and the last to be added here. Both data sheets publish the constants, the formula in both directions, and a table of code values and IRE figures at 0%, 18% and 90% reflection. All twelve of those anchors reproduce: F-Log computes 95, 469.9 and 705.4 against a published 95, 470 and 705; F-Log2 computes 95, 400.0 and 569.9 against a published 95, 400 and 570. The single F-Log2 curve covers F-Log2 C too, because Fujifilm states the two gamma curves are identical — those two differ by gamut, not by curve.

What these numbers are not

This describes encoding, not sensor performance or image quality. Code value allocation sets only how finely a signal is quantised after the sensor has already produced it. A curve that spends more code values below middle grey is not thereby cleaner in the shadows — noise comes from the sensor and the exposure. Dynamic range, colour accuracy and compression are all outside what a transfer function can tell you. And where a curve reaches full scale is a property of the equation: where a given camera actually clips is a separate, sensor-dependent number that most manufacturers do not publish.

The working behind this page is written up in how log formats allocate code values and how far apart cameras really are. For what to do with the middle-grey figure on set, see exposing S-Log3 correctly.

Published by Leumos AI. The figures come from the manufacturers' specifications, not from any Leumos measurement or product.

That product is a browser-based color grading studio, in closed beta. The waitlist is open.