为什么相机使用 Log 拍摄Why cameras shoot Log
Log 不是把画面调灰，也不是万能动态范围按钮。它是一个码值分配问题的答案：传感器以线性方式捕获 14–17 档亮度信息，而视频文件只有 10 或 12 bit 可用——Log 曲线决定每一档曝光分到多少个数字，决定哪些层次活下来、哪些被量化抹掉。这一页从数学推导、位深量化、厂商实现到后期色彩管理，把这条曲线讲透。Log isn't about making the picture grey, nor is it a magic dynamic-range button. It's the answer to a code-value allocation problem: the sensor captures 14–17 stops of luminance linearly, but a video file only has 10 or 12 bits to spend — the Log curve decides how many numbers each stop of exposure gets, which gradations survive and which get quantized away. This page works through the curve from the math, to bit-depth quantization, to vendor implementations and post-production color management.
1. 问题：线性记录把一半码值给了最亮一档
1. The problem: linear recording gives half the code values to the brightest stop传感器对光是近似线性的：进光量翻一倍（+1 档），信号数值翻一倍。这个特性在物理上很诚实，但在编码上是灾难——从剪切点往下数，第一档曝光独占全部码值的一半，第二档占四分之一，而摄影师真正在意的中灰和暗部，只能挤在剩下的零头里。A sensor responds to light roughly linearly: double the light (+1 stop) and the signal value doubles. That's physically honest but an encoding disaster — counting down from the clip point, the first stop alone takes half of all code values, the second a quarter, while the mid-grays and shadows the photographer actually cares about are crammed into the leftovers.
下面的演示按"每一档曝光分到多少个码值"统计了四种编码。点击切换，注意线性编码的柱状分布有多么倾斜，而 Log 编码几乎是平的。The demo below tallies four encodings by "how many code values each stop of exposure gets". Click to switch and notice how skewed the linear bars are, while Log is nearly flat.
光量翻倍，信号翻倍。高光天然占用更多数值空间，暗部档位被压缩到个位数码值。Double the light, double the signal. Highlights naturally hog the numeric space, while shadow stops are squeezed down to single-digit code values.
Rec.709 OETF 分配较均匀，但中灰以上约 2.5 档就到 1.0——为直出显示设计，装不下电影机的高光。The Rec.709 OETF allocates fairly evenly, but reaches 1.0 about 2.5 stops above mid-gray — designed for direct display, it can't hold a cinema camera's highlights.
纯对数区内每档分到几乎相同的码值。S-Log3 每档约 79 个 10-bit 码值，恒定不变。In the pure-log region every stop gets almost the same number of code values. S-Log3 gives about 79 ten-bit code values per stop, constant.
覆盖的档数越多，每档分到的码值越少。这就是 LogC4 要求 12 bit 的原因（见第 3 节）。The more stops you cover, the fewer code values each stop gets. That's why LogC4 requires 12 bit (see Section 3).
2. 数学：从线性、Gamma 到 Log2. The math: from linear to gamma to Log
三种映射对应三种"看待亮度"的方式。理解它们的数学差异，比背任何一条厂商曲线都更有用——因为所有 Log 曲线都是同一套思路加上不同的工程参数。Three mappings correspond to three ways of "looking at brightness". Understanding their mathematical differences is more useful than memorizing any vendor curve — because every Log curve is the same idea plus different engineering parameters.
y = k·x
等比进，等比出。+1 档输出翻倍，所以最亮一档永远占据剩余空间的一半。传感器 ADC、RAW 文件、CG 渲染都工作在这个域——适合计算，不适合存储。Ratio in, ratio out. +1 stop doubles the output, so the brightest stop always takes half the remaining space. Sensor ADCs, RAW files, and CG rendering all work in this domain — good for computation, bad for storage.
每档宽度：逐档减半Per-stop width: halves each stop
y = xγ
对数域中的直线：log y = γ·log x。它把暗部"撑开"了（γ<1 时），但每档宽度仍然不相等，且曲线必须在 x=1 处到顶——本质是为显示设备设计的，编码范围有限。A straight line in the log domain: log y = γ·log x. It "stretches" the shadows (when γ<1), but per-stop widths are still unequal, and the curve must top out at x=1 — fundamentally designed for display devices, with a limited encoding range.
每档宽度：仍逐档收缩，但比线性缓和Per-stop width: still shrinks each stop, but gentler than linear
y = m·log₂(x) + k
等比进，等差出。无论这一档在高光还是阴影，输出都增加相同的 m。这让"档"成为编码轴上的天然单位——曝光表、波形、假色、调色都能直接对话。Ratio in, equal increment out. Whether the stop is in highlights or shadows, the output rises by the same m. This makes the "stop" a natural unit on the encoding axis — light meters, waveforms, false color, and grading all speak it directly.
每档宽度：恒定 m（纯对数区内）Per-stop width: constant m (in the pure-log region)
为什么对数恰好"贴合"摄影与人眼Why the log so neatly "fits" photography and the eye
人眼：韦伯-费希纳定律指出，人对亮度的感知近似与刺激的对数成正比——我们感觉"亮了一点"对应的是亮度的固定百分比变化，不是固定增量。对数编码让每个码值步进对应大致相等的感知步长，量化误差被均匀摊开。The eye: the Weber–Fechner law says perceived brightness is roughly proportional to the log of the stimulus — "a bit brighter" corresponds to a fixed percentage change in luminance, not a fixed increment. Log encoding makes each code-value step correspond to a roughly equal perceptual step, spreading quantization error evenly.
摄影：光圈、快门、ISO、ND 全部以倍率（档）工作。负片的特性曲线（密度 vs log 曝光）本身就是对数坐标——Log 编码是这个百年传统的数字延续。Photography: aperture, shutter, ISO, and ND all work in ratios (stops). A negative's characteristic curve (density vs. log exposure) is already in log coordinates — Log encoding is the digital continuation of that century-old tradition.
信号：传感器光子散粒噪声的标准差约与信号的平方根成正比，亮部"天生"有更多噪声掩盖量化误差；对数分配把码值从噪声充裕的高光挪给干净的中暗部，恰好和噪声分布互补。The signal: the standard deviation of photon shot noise scales roughly with the square root of the signal, so highlights "naturally" have more noise to mask quantization error; log allocation moves code values away from the noisy highlights toward the cleaner mid-to-shadow region — complementing the noise distribution.
Cineon：所有现代 Log 的祖先Cineon: the ancestor of every modern Log
1990 年代初，Kodak 的 Cineon 系统要把胶片负片扫描成数字文件。胶片密度本身就是对数量，于是 Cineon 直接采用 10 bit 对数编码打印密度：黑位放在 CV 95（保留负值余量），90% 白放在 CV 685（上方留高光肩部），每 0.002 密度一个码值，约 90 个码值/档。In the early 1990s, Kodak's Cineon system had to scan film negatives into digital files. Film density is itself a logarithmic quantity, so Cineon directly used a 10-bit log encoding of printing density: black at CV 95 (leaving headroom for negatives), 90% white at CV 685 (with a highlight shoulder above), one code value per 0.002 density, about 90 code values per stop.
这套"95 / 685 / 10-bit log"的骨架影响深远：Sony 官方文档明确说 S-Log3 沿用 Cineon 式工作流习惯；DPX 文件格式、Resolve 里的 Cineon Film Log 解码，都源自这里。可以说现代相机 Log = Cineon 思路 + 各家传感器参数。This "95 / 685 / 10-bit log" skeleton runs deep: Sony's docs explicitly say S-Log3 follows Cineon-style workflow conventions; the DPX file format and Resolve's Cineon Film Log decode both trace back here. You could say modern camera Log = the Cineon idea + each vendor's sensor parameters.
x = 场景反射率（0.9 = 90% 白）；300 CV/十倍 ≈ 90 CV/档x = scene reflectance (0.9 = 90% white); 300 CV/decade ≈ 90 CV/stop
公式解剖 ①：Sony S-Log3（10 bit）Formula dissection ①: Sony S-Log3 (10 bit)
S-Log3 是公开公式中最容易读懂的一条，每个常数都有明确职责：S-Log3 is the most readable of the published formulas — every constant has a clear job:
y = ( x × (171.21 − 95) / 0.01125 + 95 ) / 1023 x < 0.01125（线性脚趾段） (linear toe segment)
420中灰锚点。Mid-gray anchor.场景 18% 灰精确落在 10-bit 码值 420（约 41% 信号电平）。波形监视器上的 S-Log3 曝光参考线就是它。 Scene 18% gray lands exactly at 10-bit code value 420 (about 41% signal level). That's the S-Log3 exposure reference line on a waveform monitor.261.5斜率 = 码值分配密度。Slope = code-value density.每十倍曝光 261.5 个码值，换算成档：261.5 × log₁₀2 ≈ 79 CV/档，全范围恒定。这一个数就决定了"每档质量"。 261.5 code values per decade of exposure; in stops: 261.5 × log₁₀2 ≈ 79 CV/stop, constant across the range. This one number sets the "quality per stop".+0.01偏置（toe 控制）。Offset (toe control).给输入加一个小常数，避免 log(0) = −∞，同时压缓极暗部的斜率，防止把传感器底噪放大记录。 Adding a small constant to the input avoids log(0) = −∞ and softens the slope in the deep shadows, preventing the sensor noise floor from being amplified into the recording.95黑位码值。Black code value.直接继承 Cineon 的 CV 95：黑不放在 0，给负向噪声和后期留缓冲。 Inherited directly from Cineon's CV 95: black isn't at 0, leaving buffer for negative noise and post.x < 0.01125线性脚趾段。Linear toe segment.极暗部改用直线连接到黑位，与对数段在拼接点斜率连续——纯对数在 0 附近会无限陡，必须截断。 The deep shadows switch to a straight line down to black, slope-continuous with the log segment at the join — a pure log is infinitely steep near 0 and must be truncated.公式解剖 ②：ARRI LogC4（12 bit）Formula dissection ②: ARRI LogC4 (12 bit)
LogC4 是更现代的设计：以 2 为底（直接对应档位）、EI 无关（任何曝光指数下解码公式相同）、面向 17 档级传感器：LogC4 is a more modern design: base-2 (mapping directly to stops), EI-independent (the decode formula is the same at any exposure index), aimed at 17-stop-class sensors:
a = (2¹⁸ − 16) / 117.45 ≈ 2231.8 b = (1023 − 95)/1023 ≈ 0.907 c = 95/1023 ≈ 0.0928
x < t 时使用线性段 (x − t)/s，与对数段在 t 处值与斜率均连续for x < t, a linear segment (x − t)/s is used, continuous in value and slope with the log segment at t
log₂以 2 为底。Base 2.x 每翻一倍（+1 档），log₂ 项加 1——档位直接成为公式的内部坐标。 Each doubling of x (+1 stop) adds 1 to the log₂ term — the stop becomes the formula's internal coordinate directly./ 1414 档对数窗口。A 14-stop log window.把 14 档的 log₂ 跨度归一化到 0–1。每档占据 b/14 ≈ 6.5% 的信号范围，即 12-bit 下约 265 CV/档。 Normalizes the 14-stop log₂ span to 0–1. Each stop occupies b/14 ≈ 6.5% of the signal range, i.e. about 265 CV/stop at 12 bit.+64, −6黑锚点。Black anchor.x = 0 时 log₂(64) = 6，减 6 后恰好为 0——纯黑精确落在编码起点，无需截断负数。 At x = 0, log₂(64) = 6, and subtracting 6 gives exactly 0 — pure black lands precisely at the encoding origin, with no need to clip negatives.a传感器满量程缩放。Sensor full-scale scaling.把相机线性信号的最大值对齐到编码窗口顶端，18% 灰则落在约 0.278（27.8%）。 Aligns the camera's max linear signal to the top of the encoding window; 18% gray then lands at about 0.278 (27.8%).b, cCineon 式装配。Cineon-style assembly.把结果装进"黑位 95、满程 1023"的传统框架——又见 95。 Packs the result into the traditional "black 95, full-scale 1023" frame — there's 95 again.对比一下斜率就明白位深要求的来源：LogC4 覆盖 14 档对数窗口，折算 10-bit 只有约 66 CV/档，低于 S-Log3 的 79——所以 ARRI 明确要求 LogC4 至少以 12 bit 编码用于采集与交换（12-bit 下为 265 CV/档）。Compare the slopes and the bit-depth requirement becomes clear: LogC4 covers a 14-stop log window, which at 10 bit is only about 66 CV/stop — below S-Log3's 79 — so ARRI explicitly requires LogC4 to be encoded at 12 bit minimum for capture and exchange (265 CV/stop at 12 bit).
为什么所有 Log 都有"脚趾"（toe）Why every Log has a "toe"
纯对数函数有两个工程缺陷：x→0 时输出趋向 −∞（数学奇点）；且在极暗部斜率无限大，意味着把传感器读出噪声和暗电流放大后写进文件。所以每条实用曲线都在暗部做处理：S-Log3 和 V-Log 用线性段拼接，LogC4 用线性段加精确的连续性约束，Log3G10 给输入加 0.01 偏置后允许负值线性延伸。A pure log has two engineering flaws: as x→0 the output goes to −∞ (a mathematical singularity), and its slope is infinite in the deep shadows, which means amplifying sensor read noise and dark current into the file. So every practical curve treats the shadows: S-Log3 and V-Log splice on a linear segment, LogC4 uses a linear segment with exact continuity constraints, and Log3G10 adds a 0.01 offset to the input and allows a linear extension into negatives.
拼接点的选择是设计决策：拼得高，暗部更"安全"但牺牲阴影层次；拼得低，阴影信息多但噪声占码值。Where to put the splice point is a design decision: splice high and the shadows are "safer" but lose gradation; splice low and there's more shadow information but noise eats code values.
为什么各家中灰落点不同Why each vendor's mid-gray lands differently
中灰放哪里 = 高光与暗部的预算分配。S-Log3 把 18% 灰放在 41%，灰上约 7.7 档到达编码上限（受 Cineon 习惯影响，暗部预算充足）；LogC4 放在 27.8%，灰上留出更多编码空间给 ALEXA 35 的高光；Log3G10 放在 1/3，明确按"灰上 10 档到 1.0"设计。Where mid-gray sits = the budget split between highlights and shadows. S-Log3 places 18% gray at 41%, reaching the encoding ceiling about 7.7 stops above gray (Cineon-influenced, with a generous shadow budget); LogC4 places it at 27.8%, leaving more encoding room above gray for the ALEXA 35's highlights; Log3G10 places it at 1/3, explicitly designed for "10 stops above gray reaches 1.0".
这就是为什么不能拿同一条监看 LUT 通吃不同机型：中灰锚点不同，LUT 输入假设就不同，画面整体明暗立即错位。This is why you can't use one monitoring LUT across different camera models: a different mid-gray anchor means a different LUT input assumption, and the whole image's brightness immediately shifts.
3. 位深与量化：banding 从哪里来3. Bit depth and quantization: where banding comes from
把连续的光编码成离散码值必然丢信息；问题是丢在哪。下面的演示用同一条"阴影渐变"（中灰下 6 档到中灰）分别走线性和 S-Log3 编码，量化到指定位深后再像后期一样提亮——线性编码的暗部断层会直接可见。Encoding continuous light into discrete code values always loses information; the question is where. The demo below runs the same "shadow ramp" (6 stops below mid-gray up to mid-gray) through linear and S-Log3 encoding, quantizes to a chosen bit depth, then lifts it like post would — the linear encoding's shadow banding becomes directly visible.
每档码值：主要曲线横向对比Code values per stop: the main curves side by side
| 曲线Curve | 容器Container | 纯对数区每档码值CV/stop in pure-log region |
|---|---|---|
| Cineon | 10 bit | ≈ 90 CV/档≈ 90 CV/stop |
| S-Log3 | 10 bit | ≈ 79 CV/档≈ 79 CV/stop |
| V-Log | 10 bit | ≈ 74 CV/档≈ 74 CV/stop |
| Log3G10 | 10 bit | ≈ 69 CV/档（高位区）≈ 69 CV/stop (upper region) |
| LogC4 | 12 bit | ≈ 265 CV/档（折算 10-bit 仅 ≈ 66）≈ 265 CV/stop (only ≈ 66 if reduced to 10-bit) |
规律：曲线想覆盖的档数越多，每档分到的码值越少。Cineon 只需服务负片密度域所以最阔绰；LogC4 要装下 17 档级传感器，10 bit 已经不够体面，必须上 12 bit。The rule: the more stops a curve covers, the fewer code values each stop gets. Cineon only had to serve the negative-density range, so it's the most generous; LogC4 must hold a 17-stop-class sensor, where 10 bit is no longer respectable, so it goes to 12 bit.
位深的实用判断Practical bit-depth judgment
8 bit（256 级）：拍 Log 时每档只剩约 20 个码值，调色一动就 banding。8 bit 机型直接拍显示 Gamma（709/HLG）反而更稳。8 bit (256 levels): shooting Log leaves only about 20 code values per stop, so the slightest grade bands. An 8-bit camera is actually safer shooting display gamma (709/HLG) directly.
10 bit（1024 级）：当前专业视频的主流底线。S-Log3、V-Log、Log3G10 都按 10 bit 设计，正常曝光 + 适度调色是安全的。10 bit (1024 levels): today's mainstream floor for professional video. S-Log3, V-Log, and Log3G10 are all designed for 10 bit, and normal exposure + moderate grading is safe.
12 bit（4096 级）：LogC4 的最低要求，也是大幅度重打光式调色的舒适区。ARRI 明确说明 10 bit 的 LogC4 不应用于归档或交换。12 bit (4096 levels): LogC4's minimum requirement, and the comfort zone for heavy relighting-style grades. ARRI explicitly states 10-bit LogC4 should not be used for archiving or exchange.
注意色度抽样：位深之外，4:2:0 对 Log 调色同样致命——绿幕、肤色微调需要 4:2:2 以上。位深管亮度阶梯，抽样管颜色分辨率，两个都要看。Mind chroma subsampling: beyond bit depth, 4:2:0 is equally fatal for Log grading — green screen and skin-tone tweaks need 4:2:2 or better. Bit depth governs the luminance staircase, subsampling governs color resolution; watch both.
4. 互动曲线：一档曝光对应多少信号4. Interactive curves: how much signal a stop of exposure maps to
拖动曝光档位或在图上滑动，读取每条曲线把"相对 18% 灰的场景亮度"映射到的信号电平。四条厂商曲线（ARRI LogC4、Sony S-Log3、Panasonic V-Log、RED Log3G10）均使用官方公开公式；可叠加 Rec.709 OETF 和 Cineon 作参照。Drag the exposure stop or slide over the chart to read the signal level each curve maps "scene brightness relative to 18% gray" to. The four vendor curves (ARRI LogC4, Sony S-Log3, Panasonic V-Log, RED Log3G10) all use the official published formulas; you can overlay the Rec.709 OETF and Cineon as references.
信号 % 为满量程归一化值，CV 为 10-bit 数据电平（信号 × 1023）。LogC4 实际应在 12-bit 容器中看待（CV × 4）。广播 IRE 与数据电平的换算：IRE ≈ (CV − 64) / 876 × 100。Signal % is the full-range normalized value; CV is the 10-bit data level (signal × 1023). LogC4 should really be viewed in a 12-bit container (CV × 4). Broadcast IRE vs. data level: IRE ≈ (CV − 64) / 876 × 100.
ARRI LogC4 的关键变化What changed with ARRI LogC4
LogC4 是 LogC3 的后继，服务 ARRI REVEAL 色彩科学与 ALEXA 35 / ALEXA 265。两个工程决定值得注意：EI 无关——LogC3 的编码随 EI 改变（每个 EI 一条曲线），LogC4 任何 EI 下编码解码公式唯一，LUT 和 IDT 管理大幅简化；12 bit 最低量化——因为 14 档对数窗口摊薄了每档码值。LogC4 配套色域为 ARRI Wide Gamut 4（AWG4，D65 白点）。LogC4 succeeds LogC3, serving ARRI REVEAL color science and the ALEXA 35 / ALEXA 265. Two engineering decisions stand out: EI-independent — LogC3's encoding changed with EI (one curve per EI), whereas LogC4 has a single encode/decode formula at any EI, greatly simplifying LUT and IDT management; and 12-bit minimum quantization — because the 14-stop log window thins out the code values per stop. Its companion gamut is ARRI Wide Gamut 4 (AWG4, D65 white point).
18% 灰 → 信号 27.8%；不要把 LogC3 的 LUT 套在 LogC4 素材上，两者中灰相差约 11 个百分点。18% gray → 27.8% signal; don't apply a LogC3 LUT to LogC4 footage — their mid-grays differ by about 11 percentage points.
S-Log3 的设计特征S-Log3's design characteristics
Sony 官方资料把 S-Log3 描述为贴近"Cineon / 负片扫描"习惯的曲线：相比 S-Log2，中灰更亮（41% vs 32%）、暗部脚趾更短、向高光延展约 1.5 档，整体更接近纯对数。搭配色域 S-Gamut3.Cine 比 DCI-P3 略宽，便于常规影视调色；S-Gamut3 则更接近传感器原生色域，留给深度色彩管理。Sony's docs describe S-Log3 as a curve close to "Cineon / negative-scan" conventions: versus S-Log2 it has a brighter mid-gray (41% vs. 32%), a shorter shadow toe, about 1.5 stops more highlight extension, and is overall closer to a pure log. Its companion gamut S-Gamut3.Cine is slightly wider than DCI-P3, convenient for regular film/TV grading; S-Gamut3 is closer to the sensor's native gamut, reserved for deep color management.
18% 灰 → CV 420（41%）；90% 白 → CV 598（61 IRE）；黑位 → CV 95。18% gray → CV 420 (41%); 90% white → CV 598 (61 IRE); black → CV 95.
5. 各家 Log 不是同一种"灰"5. Each vendor's Log isn't the same "gray"
不同厂商的 Log 围绕自家传感器、位深、监看 LUT、输出变换和色彩科学整体设计。选错任何一半（Gamma 或 Gamut），后面全错。Each vendor's Log is designed around its own sensor, bit depth, monitoring LUT, output transform, and color science as a whole. Get either half wrong (gamma or gamut) and everything downstream is wrong.
| 系统System | 常见搭配Common pairing | 设计重点Design focus | 中灰/锚点Mid-gray / anchor | 使用提醒Usage note |
|---|---|---|---|---|
| ARRI | LogC4 + ARRI Wide Gamut 4 | ALEXA 35 / 265 的 REVEAL 色彩科学；EI 无关解码；14 档对数窗口，面向 12 bit。REVEAL color science for the ALEXA 35 / 265; EI-independent decode; a 14-stop log window, aimed at 12 bit. | 18% 灰 → 27.8%；黑位 CV 95（12-bit 域为 380）。18% gray → 27.8%; black CV 95 (380 in the 12-bit domain). | LogC3 与 LogC4 不通用；需配对应 LUT、CST 或 ACES IDT。LogC3 and LogC4 aren't interchangeable; use the matching LUT, CST, or ACES IDT. |
| Sony | S-Log3 + S-Gamut3.Cine / S-Gamut3 | Cineon 式纯 Log；每档约 79 CV；S-Gamut3.Cine 面向常规调色。Cineon-style pure Log; about 79 CV/stop; S-Gamut3.Cine aimed at regular grading. | 18% 灰 → CV 420 / 41%；90% 白 → CV 598。18% gray → CV 420 / 41%; 90% white → CV 598. | 8 bit 机型拍 S-Log3 量化余量小；建议 10 bit 起，曝光偏亮保护暗部。8-bit cameras leave little quantization headroom for S-Log3; use 10 bit and up, and expose a touch bright to protect shadows. |
| Canon | Canon Log 2 / Log 3 + Cinema Gamut | Log 2 偏最大动态范围与暗部层次；Log 3 偏易用与高光延展。Log 2 leans toward maximum dynamic range and shadow detail; Log 3 toward ease of use and highlight extension. | 官方按 10% / 18% 灰给出锚点；Log 3 常见中灰约 34%。Anchors given officially against 10% / 18% gray; Log 3's mid-gray is commonly around 34%. | Log 2 后期空间大但暗部噪声更明显；一般制作 Log 3 更稳。Log 2 gives more post latitude but more visible shadow noise; for general work Log 3 is steadier. |
| Panasonic | V-Log + V-Gamut | 源自 VariCam，跨机型一致；纯对数区每档约 74 CV；V709 监看变换。Derived from VariCam, consistent across models; about 74 CV/stop in the pure-log region; V709 monitoring transform. | 18% 灰 → CV 433 / 42 IRE；90% 白 → 61 IRE。18% gray → CV 433 / 42 IRE; 90% white → 61 IRE. | V-Log L（GH/S 系微单）与完整 V-Log 编码范围不同，先确认机型。V-Log L (GH/S-series mirrorless) has a different encoding range than full V-Log; confirm the model first. |
| RED | Log3G10 + REDWideGamutRGB | IPP2 三段式管线（RAW 开发 → Log 工作空间 → 输出变换）；按"灰上 10 档"设计。The three-stage IPP2 pipeline (RAW development → Log working space → output transform); designed for "10 stops above gray". | 18% 灰 → 1/3（33.3%）；灰上 10 档 → 1.0。18% gray → 1/3 (33.3%); 10 stops above gray → 1.0. | 区分 Legacy 与 IPP2；输出变换是显示适配的一部分，不是装饰 LUT。Distinguish Legacy from IPP2; the output transform is part of display adaptation, not a decorative LUT. |
| Blackmagic | BMD Film Gen 5 / DaVinci Wide Gamut | 围绕 Resolve 色彩管理与 BRAW 元数据；常转入 DWG/Intermediate 再调色。Built around Resolve color management and BRAW metadata; usually converted into DWG/Intermediate before grading. | 不同机型与 Gen 版本各异。Varies by model and Gen version. | 确认 Gen 4 / Gen 5、BMD Film 与 DWG Intermediate，别只看 "Film" 字样。Confirm Gen 4 / Gen 5, BMD Film vs. DWG Intermediate — don't just go by the word "Film". |
6. Log vs RAW vs 显示 Gamma：三种记录哲学6. Log vs. RAW vs. display gamma: three recording philosophies
这三者不是"画质从高到低"，而是把哪些决定烘焙进文件、把哪些留给后期的三种取舍。Log 的位置很特别：它已经是"显像后"的视频，却尽力保留"显像前"的调整空间。These three aren't a "best-to-worst image quality" ranking, but three trade-offs over which decisions get baked into the file and which are left for post. Log's position is special: it's already "developed" video, yet it tries hard to preserve "pre-development" latitude.
| 维度Dimension | RAW | Log（10/12 bit）Log (10/12 bit) | 显示 Gamma（709 / HLG）Display gamma (709 / HLG) |
|---|---|---|---|
| 数据形态Data form | 传感器马赛克线性数据，未去拜耳Linear sensor mosaic data, not debayered | 已去拜耳的三通道视频，亮度经 Log 编码进专业编解码器Debayered three-channel video, luminance Log-encoded into a professional codec | 已烘焙显示曲线与色域的成品视频Finished video with the display curve and gamut already baked in |
| 白平衡 / ISOWhite balance / ISO | 多为元数据，后期近乎无损改动Mostly metadata, changeable in post nearly losslessly | 已烘焙进画面，后期只能近似补偿Baked into the image; post can only approximate a correction | 完全烘焙，改动即画质损失Fully baked; any change costs image quality |
| 位深与体积Bit depth & size | 12–16 bit，体积最大（压缩 RAW 可缓解）12–16 bit, largest size (compressed RAW eases this) | 10–12 bit，ProRes / XAVC / BRAW 级体积10–12 bit, ProRes / XAVC / BRAW-class size | 8–10 bit，体积最小8–10 bit, smallest size |
| 动态范围存留Dynamic range retained | 传感器全部（受 RAW 压缩比影响极小）All of the sensor's (barely affected by RAW compression ratio) | 接近传感器全部——前提是曝光正确Close to all of the sensor's — provided exposure is correct | 受显示曲线剪切，高光最先丢Clipped by the display curve; highlights go first |
| 后期成本Post cost | 最高：RAW 开发 + 色彩管理 + 调色Highest: RAW development + color management + grading | 中：色彩管理 + 调色Medium: color management + grading | 最低：可直接剪辑出片Lowest: edit and deliver directly |
| 适合场景Best for | 电影、广告、重后期、视效素材Film, commercials, heavy post, VFX plates | 绝大多数专业视频制作的甜点区The sweet spot for the vast majority of professional video | 新闻、直播、活动记录、快速交付News, live broadcast, event coverage, fast delivery |
"部分开发"的 RAW"Partially developed" RAW
BRAW 和 ProRes RAW 在机内做了部分处理（降噪、部分去拜耳运算），严格说是"半开发 RAW"。白平衡 / ISO 仍以元数据方式可调，但不要按"完全未处理"理解它们。BRAW and ProRes RAW do some in-camera processing (noise reduction, partial debayer math), so strictly they're "semi-developed RAW". White balance / ISO are still metadata-adjustable, but don't think of them as "completely unprocessed".
RAW 里也有 LogThere's Log inside RAW too
很多压缩 RAW 在存储层内部使用对数或类对数量化来省位深——这与"拍 Log 视频"是两回事。前者是文件内部实现，后者是交付给后期的色彩空间约定。Many compressed RAW formats use log or log-like quantization internally at the storage layer to save bits — that's a different thing from "shooting Log video". The former is a file-internal implementation; the latter is a color-space agreement handed to post.
Log 是性价比拐点Log is the value sweet spot
从显示 Gamma 升级到 Log，调色空间的提升立竿见影；从 Log 升级到 RAW，提升集中在白平衡反悔权与极端调色场景。预算和存储有限时，10 bit Log + 正确曝光是收益最大的一档投入。Upgrading from display gamma to Log gives an immediate jump in grading latitude; upgrading from Log to RAW mostly adds white-balance undo rights and extreme-grade cases. When budget and storage are limited, 10-bit Log + correct exposure is the highest-return level of investment.
7. 后期工作流：把 Log 正确"翻译"回去7. Post workflow: "translating" Log back correctly
Log 素材的一切价值都要靠正确的色彩管理兑现。这一节讲清三个工具的本质区别（CST / LUT / ACES）、Resolve 的两种标准做法，以及一张常见故障诊断表。All the value of Log footage is realized only through correct color management. This section clarifies the essential differences between three tools (CST / LUT / ACES), two standard Resolve approaches, and a common-fault diagnostic table.
解析公式，精确可逆Analytic formula, precisely invertible
Color Space Transform 用厂商公式做数学变换，没有网格插值误差，超出范围的值可保留（取决于设置），方向可逆。Resolve 的 CST 节点和色彩管理内部都是这种方式。A Color Space Transform does the math with the vendor formula, with no grid-interpolation error; out-of-range values can be preserved (depending on settings), and it's reversible. Resolve's CST node and its internal color management both work this way.
适合：技术转换（Log→工作空间→交付空间）的全部环节。Good for: every stage of technical conversion (Log → working space → delivery space).
采样网格，近似固化A sampled grid, baked approximation
LUT 把变换烘焙成 17³/33³/65³ 网格点，运行时插值。输入域之外的值会被剪切，精度受网格密度限制，且不可逆。优势是能固化任意复杂的创意外观、跨软硬件交换（监视器、交换给剪辑）。A LUT bakes the transform into 17³/33³/65³ grid points and interpolates at runtime. Values outside the input domain get clipped, precision is limited by grid density, and it's not reversible. Its strength is freezing an arbitrarily complex creative look and exchanging it across software/hardware (monitors, handing off to editorial).
适合：现场监看、创意外观分发；不适合：当作万能技术转换。Good for: on-set monitoring and distributing a creative look; not for: serving as a universal technical conversion.
标准化框架，不是工具A standardized framework, not a tool
ACES 规定一条固定管线：相机素材经 IDT 进入场景线性的 ACES2065-1（调色常用 ACEScct），输出经 RRT + ODT 适配显示。价值在于多机型统一、跨机构交换和长期归档，而非"更好看"。ACES prescribes a fixed pipeline: camera footage enters scene-linear ACES2065-1 via an IDT (ACEScct is common for grading), and the output adapts to the display via RRT + ODT. Its value is unifying many camera models, cross-facility exchange, and long-term archiving — not "looking better".
适合：多机位混拍、视效协作、需要归档交换的项目。Good for: multi-camera shoots, VFX collaboration, and projects that need archival exchange.
Resolve 做法 A：自动色彩管理（推荐起点）Resolve approach A: automatic color management (recommended starting point)
- 项目设置 → 色彩科学选
DaVinci YRGB Color Managed，模式选DaVinci Wide Gamut Intermediate。Project settings → set Color Science toDaVinci YRGB Color Managedand the mode toDaVinci Wide Gamut Intermediate. - 让 Resolve 按剪辑元数据自动识别每条素材的输入空间；无元数据的素材，右键 →
Input Color Space手动指定（Gamma 和 Gamut 一起选）。Let Resolve auto-detect each clip's input space from clip metadata; for clips without metadata, right-click →Input Color Spaceand set it manually (choose gamma and gamut together). - 时间线在宽色域中间空间调色；输出色彩空间按交付目标设
Rec.709 Gamma 2.4/ P3 / PQ。Grade the timeline in the wide-gamut intermediate space; set the output color space toRec.709 Gamma 2.4/ P3 / PQ per the delivery target. - 检查点：拉一条曝光正常的素材，中灰应落在波形约 38–45%（取决于输出曲线），肤色不偏——否则九成是输入空间错了。Checkpoint: pull up a correctly exposed clip — mid-gray should land around 38–45% on the waveform (depending on the output curve) and skin tones shouldn't shift; otherwise the input space is wrong nine times out of ten.
Resolve 做法 B：手动 CST 节点（精确可控）Resolve approach B: manual CST node (precise control)
- 保持项目为
DaVinci YRGB（非托管），时间线色彩空间设为 DWG/Intermediate。Keep the project onDaVinci YRGB(unmanaged), and set the timeline color space to DWG/Intermediate. - 每条素材第一个节点放 CST：
输入 = 素材实际 Gamma/Gamut→输出 = DWG / Intermediate。Put a CST on each clip's first node:input = the clip's actual gamma/gamut→output = DWG / Intermediate. - 中间节点自由调色（此时工作在统一的宽色域空间，工具行为跨机型一致）。Grade freely on the middle nodes (now working in one unified wide-gamut space, so tool behavior is consistent across camera models).
- 节点链最末端放输出 CST：
DWG → Rec.709 / Gamma 2.4（或目标交付空间），并打开 Tone/Gamut Mapping 处理超范围值。Put the output CST at the very end of the node chain:DWG → Rec.709 / Gamma 2.4(or the target delivery space), and enable Tone/Gamut Mapping to handle out-of-range values.
故障诊断表：症状 → 原因 → 修复Fault diagnostic table: symptom → cause → fix
| 症状Symptom | 最可能原因Most likely cause | 修复Fix |
|---|---|---|
| 画面整体发灰、对比饱和都低Whole image looks gray, low contrast and saturation | 直接在看 Log 原始信号，没有做任何输出/监看变换You're viewing the raw Log signal directly, with no output/monitoring transform | 启用色彩管理或在末端加 CST/官方 LUT；Log 素材"看起来灰"是正常的中间态Enable color management or add a CST/official LUT at the end; Log footage "looking gray" is a normal intermediate state |
| 暗部死黑、高光断层、肤色过饱和Crushed shadows, broken highlights, oversaturated skin | 双重变换：素材已在机内烘焙了 709（或已带 LUT），又叠了一次 Log→709Double transform: the footage already baked in 709 in-camera (or carries a LUT), and you added another Log→709 | 用示波器确认素材实际状态（Log 素材波形通常集中在 20–60%），删掉多余的一次变换Use a scope to confirm the footage's actual state (Log waveforms usually sit around 20–60%) and remove the redundant transform |
| 整体明暗错位，波形整体偏高或偏低Overall brightness off, waveform sits too high or too low | 输入 Gamma 选错：S-Log2 当 S-Log3、LogC3 当 LogC4、V-Log L 当 V-LogWrong input gamma: S-Log2 read as S-Log3, LogC3 as LogC4, V-Log L as V-Log | 核对拍摄元数据和场记，按机型与固件确认实际记录曲线Check the shoot metadata and camera reports, confirming the actual recording curve by model and firmware |
| 亮度正常但颜色不对，红绿色相偏移Brightness fine but colors wrong, red/green hue shift | Gamut 选错：只对了 Gamma，Color Space 用了默认或别家Wrong gamut: only the gamma is right, while the color space is left at default or another vendor's | 输入变换同时指定 Gamut（S-Gamut3.Cine、AWG4、V-Gamut、RWG 等）Specify the gamut in the input transform too (S-Gamut3.Cine, AWG4, V-Gamut, RWG, etc.) |
| 暗部一提亮就出现条带 bandingShadows band as soon as you lift them | 8 bit（或重压缩）记录的 Log，或严重曝光不足后强提Log recorded at 8 bit (or heavily compressed), or a hard lift after severe underexposure | 至少 10 bit 记录；现场按厂商建议曝光（多数 Log 建议偏亮 0.5–1 档保护暗部）Record at 10 bit minimum; expose per the vendor's recommendation on set (most Logs suggest 0.5–1 stop brighter to protect shadows) |
| 套用别人的 LUT 完全不对劲Someone else's LUT looks completely wrong | LUT 的输入假设不同：机型、Gamma、Gamut 或电平范围（Legal/Full）不匹配The LUT's input assumptions differ: camera model, gamma, gamut, or level range (Legal/Full) don't match | 查明 LUT 期望的输入空间，先用 CST 把素材转换到该空间，再挂 LUTFind out the LUT's expected input space, use a CST to convert the footage into that space first, then apply the LUT |
| 高光一调就剪切，信息找不回来Highlights clip the moment you grade them, with no recovery | LUT 或输出变换放在节点链太靠前，后续调整工作在被剪切的显示域The LUT or output transform sits too early in the node chain, so later adjustments work in the already-clipped display domain | 曝光、白平衡等"场景域"调整放在输出变换之前；LUT 永远放链路末端Put "scene-domain" adjustments like exposure and white balance before the output transform; always put the LUT at the end of the chain |
按锚点曝光Expose to the anchor
用波形对中灰锚点（S-Log3 看 41%、V-Log 看 42%、LogC4 看 28%），高光看斑马/假色。不要赌"后期拉得回来"。Use the waveform against the mid-gray anchor (41% for S-Log3, 42% for V-Log, 28% for LogC4) and zebras/false color for highlights. Don't gamble on "fixing it in post".
LUT 只管显示The LUT is display-only
现场监看 LUT 是显示变换，不写入素材。给导演看正常画面，机器记录 Log 原文。An on-set monitoring LUT is a display transform, not written into the footage. Show the director a normal picture while the camera records the Log original.
先技术后创意Technical before creative
先把输入空间认对、转进工作空间，再开始创意调色。顺序反了，所有调整都建立在错误地基上。Get the input space right and convert into the working space first, then start the creative grade. Reverse the order and every adjustment is built on a wrong foundation.
按交付适配Adapt to the delivery
Rec.709、P3、PQ、HLG 各有输出变换。最终观感由输出端决定，Log 只是中间载体。Rec.709, P3, PQ, and HLG each have their own output transform. The final look is decided at the output end; Log is only the intermediate carrier.
8. 常见误区8. Common misconceptions
这几句话流传很广，每一条都把 Log 的某个侧面绝对化了。These lines circulate widely, and each one takes one aspect of Log and makes it absolute.
❌ "拍 Log = 动态范围更大""Shooting Log = more dynamic range"
动态范围由传感器决定，Log 只决定编码质量。Dynamic range is set by the sensor; Log only decides encoding quality.同一台相机拍 Log 和拍 709，传感器捕获的档数相同；区别是 709 在编码时把装不下的高光剪掉了，而 Log 把它们压进了文件。Log 保全，而非创造。 The same camera captures the same number of stops whether shooting Log or 709; the difference is that 709 clips the highlights it can't hold during encoding, while Log compresses them into the file. Log preserves, it doesn't create.
❌ "Log 素材必须看起来很灰才专业""Log footage has to look gray to be professional"
灰是中间态，不是目标。Gray is an intermediate state, not the goal.Log 画面灰是因为对数信号直接被显示器当作显示信号解读。正确流程永远配监看/输出变换。以"灰"为美开拍、不做变换直接交付，是对 Log 的双重误用。 Log looks gray because the log signal is read directly by the monitor as a display signal. A correct workflow always pairs it with a monitoring/output transform. Treating "gray" as the aesthetic while shooting, and delivering without a transform, is a double misuse of Log.
❌ "一个 LUT 可以通吃所有机型""One LUT works for every camera"
LUT 是针对特定输入空间的函数。A LUT is a function for a specific input space.各家中灰锚点（28%–42%）和色域都不同，同一 LUT 喂不同 Log，明暗和色彩立即错位。跨机型统一的正确做法是先 CST 到共同工作空间。 Each vendor's mid-gray anchor (28%–42%) and gamut differ, so feeding the same LUT different Logs immediately throws off brightness and color. The right way to unify across cameras is to CST into a common working space first.
❌ "8 bit 也能拍 Log，反正后期能救""8 bit can shoot Log too, post will save it"
8 bit 拍 Log 每档只剩约 20 个码值。8-bit Log leaves only about 20 code values per stop.轻度调色或许侥幸，提亮暗部、拉肤色立刻 banding。8 bit 机器更稳的选择是拍显示 Gamma 或 HLG，把量化预算用在最终影调上。 A light grade might get away with it, but lifting shadows or pulling skin tones bands immediately. An 8-bit camera's safer choice is to shoot display gamma or HLG, spending the quantization budget on the final tonality.
资料来源Sources
本页曲线全部使用厂商公开公式实现；常数与锚点可在下列文档中核对。Every curve on this page is implemented from the vendors' published formulas; the constants and anchors can be verified in the documents below.