How Photography Works | Bit Depth and Banding — Why Smooth Gradients Break Into Steps When Too Few Levels Survive

A blue sky can be physically smooth and still break into visible steps on the screen.

The atmosphere did not suddenly form stripes. The representation ran out of sufficiently fine numerical levels, or processing stretched the available levels until the gaps became visible.

Bit depth determines how many discrete numerical levels are available. Banding appears when those discrete steps become visible instead of reading as a smooth gradient.

This article continues the canonical How Photography Works | Every Photograph Leaves Something Out knowledge map. Colour Spaces and Gamut owns how large a colour territory a representation can cover. Bit Depth owns how finely that territory can be divided into distinct encoded values.

Quick Read

Bit depth sets the number of possible code values available per channel or per sample. An 8-bit channel provides 256 numerical levels; higher bit depths provide many more. More levels do not automatically mean more visible colours in every final image, but they give processing more numerical precision and reduce the chance that strong edits will reveal gaps between adjacent values. Banding or posterisation occurs when smooth source variation is represented by too few distinct output levels. Noise or intentional dithering can sometimes make the steps less visible by distributing quantisation error spatially.

continuous scene variation → finite code values → quantisation → output gradient

Digital Images Cannot Store Infinite Gradations

The physical world can vary continuously over ranges relevant to photography. Digital storage represents those measurements with finite numbers. Every stored value therefore represents an interval of possible underlying values.

This approximation is called quantisation. With enough levels relative to the signal and viewing conditions, the steps become effectively invisible. With too few, the staircase emerges.

Eight Bits Means 256 Code Values Per Channel

An unsigned 8-bit channel can represent integer values from 0 to 255. In an RGB image that allows millions of possible triplets, but those combinations are not distributed uniformly in perceptual colour or luminance.

The important point is that each channel has finite precision. There is no stored value between 127 and 128 in that encoded channel.

Sixteen-Bit Workflows Provide Much Finer Numerical Steps

A 16-bit integer channel can provide up to 65,536 code values. Many photographic pipelines use lower effective sensor precision or specialised packed formats, but high-bit-depth working files still provide far more room for intermediate calculations than 8-bit output.

This matters especially during strong tonal, colour and local adjustments where values are repeatedly transformed and redistributed.

More Bit Depth Does Not Mean the Sensor Measured 65,536 Perfectly Distinct Tones

Stored precision and measurement precision are different. Noise may be larger than the difference between neighbouring code values. A nominally high-bit-depth file can contain many numerical steps that do not correspond to equally precise scene distinctions.

This connects to Noise. Extra code values can preserve arithmetic headroom without creating certainty the sensor never had.

Banding Is Quantisation Made Visible

A smooth sky ideally changes gradually from one blue value to another. If the output can use only a small number of distinct values across that range, neighbouring pixels are forced into the same codes. Broad regions become identical until the next code threshold is crossed.

The eye sees those thresholds as bands.

Posterisation Is the More General Failure

Posterisation describes tonal or colour regions collapsing into visibly discrete levels rather than smooth transitions. Banding is a common form of posterisation in gradients, but the same principle can affect skin, shadows, fog and colour transitions.

The name comes from the flat stepped appearance associated with limited-tone poster printing.

Strong Editing Can Expose Steps That Were Hidden

Suppose an 8-bit gradient uses only values 100 through 120. A strong contrast expansion stretches those 21 codes across a much larger visible output range. What used to be tiny numerical steps become large tonal jumps.

This is why high-bit-depth editing is useful even if the final delivery file is 8-bit. The intermediate calculations can preserve much finer distinctions until the last conversion.

Tone Curves Can Amplify Quantisation

A steep section of a tone curve expands differences in one tonal region. If only a few input codes occupy that region, the curve can turn small code steps into visible bands.

Tone Curves therefore operate more gracefully when adequate numerical precision survives upstream.

White Balance Can Stretch One Channel More Than Another

Correcting a strongly warm or cool image may require multiplying one colour channel substantially. Weakly populated values in that channel can become more widely separated after the transform.

Again, RAW and higher-bit-depth workflows provide more room for the transformation than a heavily processed 8-bit derivative.

Wide Gamut and Low Bit Depth Can Be an Awkward Combination

A wider colour space covers a larger region. If the same limited number of code values is spread over that larger region, numerical steps can represent larger colour differences in some areas.

This is why Colour Spaces and Gamut and bit depth belong together: range and precision must be balanced.

Gamma Encoding Changes Where the Steps Are Spent

Common image encodings do not allocate numerical codes linearly to physical luminance. Nonlinear transfer functions give more code precision to darker perceptual regions than a simple linear 8-bit encoding would.

This helps make 8-bit delivery surprisingly effective under ordinary viewing conditions. The encoding is designed around perceptual efficiency, not equal physical-light increments.

A Perfect Gradient Can Band on the Display

The source file may contain sufficient precision, but the graphics pipeline, display panel, operating system or connection may reduce effective bit depth. A monitor operating at limited precision can reveal bands that are not strongly present in the file.

The receiver is part of the system. A 16-bit master cannot force a low-precision display to show every distinction.

Compression Can Add Another Quantisation Layer

JPEG compression quantises transformed image coefficients. Smooth gradients can therefore accumulate both ordinary code-value limitations and compression artefacts.

JPEG Compression owns that transformed-domain loss.

Dithering Uses Noise to Hide Steps

Add a tiny structured or random variation around a quantisation boundary and neighbouring pixels do not all jump to the next code at the same location. The transition becomes spatially distributed rather than forming one clean band.

The image contains more noise but can look smoother. This is a beautiful example of adding uncertainty to improve perceptual continuity.

Noise Can Naturally Mask Banding

A perfectly smooth digital gradient is more likely to reveal code boundaries than a naturally noisy photograph. Sensor noise, texture and grain-like variation can break up the visual coherence of the bands.

Denoising an already low-bit-depth sky too aggressively can therefore make banding more visible by removing the variation that was hiding the steps.

Banding Can Also Come From Uneven Illumination

Not every visible band is quantisation. Flickering lights, rolling shutter, sensor readout interactions and display refresh can produce horizontal or vertical bands with a different physical cause.

Diagnosis matters. Quantisation banding follows tonal steps; illumination banding follows timing and source variation.

Bit Depth Does Not Equal Dynamic Range

Dynamic range describes the ratio between the largest usable signal and the smallest usable signal above the noise floor. Bit depth describes how many digital codes are available to represent values.

A system can have many code values but poor dynamic range because noise dominates the lower end. It can also have meaningful dynamic range that is then encoded inefficiently into too few output levels.

Bit Depth Does Not Equal Colour Accuracy

More numerical levels allow finer representation, but they do not guarantee correct colour. A badly calibrated pipeline can represent the wrong colour very precisely.

Accuracy depends on calibration, profiles, illumination, sensor response and conversion as well as code precision.

RAW Bit Depth and Output Bit Depth Serve Different Stages

A camera may record 12-bit, 14-bit or another sensor data precision internally, while the final web JPEG is 8-bit per channel. The high-precision capture supports flexible processing; the lower-precision delivery is often sufficient after the final tone and colour mapping are fixed.

Reducing bit depth late is very different from starting every edit with a heavily quantised derivative.

The Final Viewer Often Receives Eight Bits—and That Can Be Enough

For well-prepared photographs displayed under ordinary conditions, 8-bit-per-channel output can look beautifully smooth because perceptual encoding, texture, display behaviour and viewing distance hide the discrete levels.

High bit depth is not about proving that 8-bit images are bad. It is about protecting the image while it is still being transformed.

Scientific Images Need Quantisation Awareness

If pixel intensity is used quantitatively, reducing bit depth can merge distinct measurements into the same code. A display version may be adequate for visual inspection while the original high-precision data remains necessary for measurement.

Scientific Photography Is Measurement With a Camera therefore separates analytical data from reduced visual derivatives.

The Bit-Depth and Banding Audit

1. Capture precision: what bit depth or effective sensor precision exists upstream?
2. Working precision: is editing occurring in 8-bit or higher precision?
3. Gradient: which smooth tonal or colour regions are vulnerable?
4. Transform: are curves, white balance or saturation stretching limited values?
5. Gamut: is a wide colour range being encoded with sufficient precision?
6. Compression: has JPEG or another lossy stage added quantisation?
7. Denoising: has natural variation that masked steps been removed?
8. Dither: could controlled noise reduce visible banding?
9. Display: is the output device itself limiting precision?
10. Purpose: visual delivery, archive, heavy editing or quantitative measurement?

Photography Laboratory 1: Gradient Stress Test

Create or photograph a smooth gradient and make high- and low-bit-depth working copies if your software supports them. Apply strong curves and saturation changes, then compare the onset of banding.

Photography Laboratory 2: Dither the Gradient

Take a visibly banded test gradient and add a very small amount of fine noise. Compare the visible boundaries. Observe how added variation can make the output look smoother without increasing the number of code levels.

Photography Laboratory 3: Edit Before and After 8-Bit Conversion

From the same high-quality master, perform a strong tonal edit once before 8-bit conversion and once after making an 8-bit derivative. Compare smooth skies, shadows and colour transitions.

For Primary Readers

Imagine trying to draw a sunset with only five crayons from dark blue to light blue. You will see steps. Give yourself fifty shades and the transition can look much smoother.

For Secondary Readers

Connect bit depth to powers of two and quantisation. Explain why 8 bits gives 256 code values and why stretching a small subset of those values can reveal visible gaps.

For Advanced Readers

Analyse quantisation error, code-value allocation under nonlinear transfer functions, effective number of bits, dithering, noise shaping and cumulative precision loss through colour transforms, resampling and lossy compression. Distinguish nominal storage precision from information capacity set by sensor noise and downstream devices.

Common Misconceptions

• “16-bit means the camera measured 65,536 perfectly accurate tones.” Storage precision can exceed measurement precision.
• “Banding always means low bit depth.” Lighting flicker and sensor timing can create different kinds of bands.
• “More bit depth means wider gamut.” Bit depth is precision; gamut is representable range.
• “8-bit photographs are low quality.” Well-rendered 8-bit output can be excellent; higher precision is especially valuable during processing.
• “Adding noise always worsens an image.” Fine dithering can reduce visible quantisation steps.

Frequently Asked Questions

Why does banding appear after editing?

Strong edits can expand a small set of code values across a wider output range, making the gaps between those values visible.

Why can noise hide banding?

Random variation distributes transitions across neighbouring pixels instead of allowing an entire region to change code at the same boundary.

Do I need 16-bit files for the web?

Usually not for final delivery. Higher precision is most useful while editing; final web output is commonly converted to a suitable 8-bit delivery format.

Final Thought: Smoothness Is Built From Enough Small Steps

A digital photograph is never infinitely continuous. It succeeds because the steps are usually small enough, well allocated enough and noisy enough that our eyes experience continuity instead of a staircase.

Bit depth is not the beauty of the gradient. It is the numerical room that lets the gradient survive everything we do to it before someone finally sees it.