Exposure first. Encoding second.

A scene can contain a bright window and dark objects at once. Light ratios are conveniently described in stops: one stop doubles or halves exposure. In linear integer encoding, the top half of the range receives as many codes as all lower stops combined. This is not a defect of linear light; it is a question of distributing finitely many levels.

Log encoding reallocates those levels to make a large relative range convenient to store and process. It does not require immediately fitting the scene to one display contrast. Retained detail still depends on the complete acquisition chain: sensor, exposure, noise, encoding and compression. Not every display curve necessarily clips highlights, and the label log alone does not guarantee their survival.

A pure logarithm turns multiplication into addition: log₂(2x) = log₂(x) + 1. Equal stops therefore occupy similar code intervals in its logarithmic region. Actual camera curves include scale, offsets and piecewise sections. Black need not have code zero, and an 18% reference need not encode to 0.18.

The lab implements ARRI’s software functions in specification §4.1. Threshold t separates the logarithmic branch from a linear extension for negative values introduced during processing. This is not the sensor hardware curve in §3. The inverse function first restores relative linear values; it performs neither a gamut transform nor a complete rendering for a monitor.

Log numbers describe a scene-relative signal, not the display codes expected for your final image. Direct viewing interprets midtones, black and highlights incorrectly. Misassigned primaries can alter colour relationships too. Raising Contrast and Saturation may produce a pleasing image without proving the transform is correct.

Not every flat clip is log, however. A Video/Full mismatch, a baked soft grade or incorrect HDR viewing can look similar. Conversely, a log file can appear contrasty because a monitoring LUT is enabled. Establish what the file contains first, then which transforms affect the Viewer.

A transfer function relates linear values to codes; it does not define primary chromaticities. “S-Log3” is therefore incomplete without S-Gamut3 or S-Gamut3.Cine. ARRI generations of LogC and Wide Gamut must also be distinguished rather than mixed by familiar names.

An input transform must match the actual signal arriving at it. For RAW, that may be the chosen decoder output rather than the original sensor representation. Record gamut, curve and processing version first, then check whether colour management already performs the input transform. Transforming correctly decoded footage twice creates a new error rather than greater accuracy.

LogC4 combines a defined transfer curve with ARRI Wide Gamut 4. The gamut specifies its primaries and D65; the curve represents relative scene values. Software LogC4 decoding does not require selecting a different curve shape for each EI. This does not make exposure or EI irrelevant to acquisition and headroom.

The specification includes an important constraint: LogC4 was designed for at least 12-bit quantisation, and ARRI explicitly discourages 10-bit LogC4 for archive or interchange. The lab therefore labels its 8/10-bit options as loss demonstrations, not production recording recommendations. LogC4 cannot be decoded as LogC3 simply because the names are similar.

Sony S-Log3 can accompany different sets of primaries. S-Gamut3 and S-Gamut3.Cine are not two saturation strengths: they are different RGB definitions requiring matching transforms. The Cine suffix does not mean that a file is already DCI-P3 or ready for theatrical projection.

Sony’s technical summary describes S-Gamut3.Cine as a cinema-oriented gamut encompassing P3 without being identical to it. The practical lesson is not to choose a “better” triangle, but to retain the recording-mode information. A historical specification table should not be treated as a list of currently supported cameras; verify each device separately.

Blackmagic Film and DaVinci Intermediate serve different purposes. Film describes a camera-oriented nonlinear representation for its colour-science generation; Intermediate accompanies DaVinci Wide Gamut as a working representation. They must not be substituted as Input Gamma just to obtain a similar contrast.

BRAW retains data and metadata for controlled decoding. Decode settings can change the RGB delivered to later tools. The current decoder output therefore matters as well as the original camera mode. Available controls depend on format and version; inspect the actual clip. This lab implements neither the Film Gen 5 curve nor the BRAW SDK rendering.

RAW describes a class of retained data and deferred processing decisions. Log describes value encoding. RAW files can also use nonlinear packing, and decoding can produce log RGB. Thus “RAW is always linear and log is always non-RAW” is too crude.

An ordinary recorded log video clip has already undergone some camera processing and does not become RAW by assigning another gamma. Post-capture white-balance flexibility depends on retained information and decoder support. BRAW performs part of its demosaic in camera, illustrating why all RAW formats cannot be assigned one identical processing diagram.

A technical transform starts from a known source state. Managed projects use RCM or ACES rules; explicit workflows use a correctly configured transform. Its destination can be a working representation, not immediately Rec.709. Output rendering then prepares the result for the target display.

Balance and exposure operations can occupy different positions in a properly designed workflow. What matters is knowing which values each tool expects. It is incorrect to insist that all creative processing must follow a display LUT. A viewing LUT and a LUT baked into pixels are different states too. Bypass stages one at a time on a test clip to locate duplicate input or output transforms.

Log cannot repair a saturated photosite or retroactively improve signal-to-noise ratio. Brightening an underexposed take lifts noise alongside signal. Darkening a clipped area can make it grey without restoring distinctions that were never measured.

“Overexpose every log camera by two stops” is not a reliable rule. Exposure decisions depend on camera, mode, noise, motion and important highlights. EI/ISO can affect analogue gain, metadata and headroom allocation differently. The lab studies only encoding: its slider is neither an ISO model nor a real sensor-noise simulation.