Overview
"16-bit" refers to a color depth (also bit depth or quantization depth) of 16 bits per color channel. This provides 2^16 = 65,536 discrete tonal steps per channel. For a three-channel RGB image, this mathematically results in approximately 281 trillion representable colors (65,536^3). More bits per channel mean finer gradations between adjacent brightness values, thus less visible banding in gradients, especially during aggressive color correction.
Important: Bit depth is not the same as dynamic range. Higher bit depth does not create additional exposure latitude; it only determines how finely the existing tonal range is sampled (quantized). How efficiently the steps are utilized also depends on the encoding (see below).
Integer vs. Float
"16-bit" is not unambiguous as long as the number format remains unspecified. In practice, two variants occur:
- 16-bit Integer: whole number values from 0 to 65,535. Typical for RAW photos and the 16-bit mode of some image programs (e.g., Adobe After Effects' 16-bit mode operates with integers).
- 16-bit Half-Float (Half-Precision): floating-point number according to IEEE 754. The container format OpenEXR, developed by Industrial Light & Magic for the VFX pipeline, stores Half-Float and can therefore store values above white ("super white") as well as very fine shadow differences in a scene-referred linear manner.
In VFX and compositing, 16-bit Half-Float is considered the standard for visual image channels (color, light), while 32-bit Float is reserved for data passes such as depth, position, or motion vector channels that require higher precision.
In Camera and RAW Context
For camera sensors, bit depth describes the resolution of the Analog-to-Digital Converter (ADC) or internal processing. Manufacturers often process with higher internal bit depth than they record finally:
- ARRI: internally processes images from most digital cameras in 16-bit linear and records them as 12-bit Log ARRIRAW; the ALEXA 35 processes in 18-bit linear and stores 13-bit Log.
- RED: advertises "16-bit RAW"; the recorded REDCODE practically operates with a lower effective bit depth.
The background is the encoding: linear storage distributes code values unevenly (the brightest stops receive the most steps), while logarithmic encoding (e.g., LogC) assigns approximately the same number of steps to each stop. Therefore, in Log, fewer bits are often sufficient for a visually equivalent result.