Why 10-Bit Video Helps, but Does Not Guarantee Smooth Gradients

Watch a sunset, a dark sky, a fog bank, or a slow fade.

The image should change smoothly.

Sometimes it breaks into visible steps. Large regions become slightly different shades, with contour lines between them.

That artifact is called banding or false contouring.

More bit depth can reduce the risk because it gives the signal more available code values. That is why ten-bit representation is important to HDR.

It is not the only part of the answer.

A ten-bit signal can already contain banding. An eight-bit signal can look smooth. A native ten-bit panel can reveal damaged gradients, and a display using dithering can sometimes hide them.

The practical answer is:

1. Preserve the source's intended bit depth and signal range.
2. Do not assume a ten-bit label proves ten-bit precision through the entire chain.
3. Treat compression, graphics rendering, color conversion, tone mapping, and display processing as possible causes.
4. Enable the input mode required for the source format when the television's manual calls for it.
5. Use a certified HDMI cable appropriate to the required bandwidth.
6. Do not expect a cable or picture setting to restore gradient levels that were lost earlier.
7. Use test patterns and source comparisons to locate the stage that introduced the contour.

What bit depth counts

A bit has two possible states.

Each added bit doubles the number of mathematical values.

These are calculated as:

• 2^8 = 256
• 2^10 = 1,024
• 2^12 = 4,096

The values are often written:

• 0 through 255
• 0 through 1,023
• 0 through 4,095

That notation describes the complete unsigned integer range.

Television does not always use the complete range for nominal picture black through nominal picture peak.

Mathematical range is not the same as nominal video range

Recommendation ITU-R BT.2100-3 defines both narrow and full-range 10-bit and 12-bit HDR representations. ^[1]

For achromatic or luma-like components:

Narrow range is the default programme-exchange representation in BT.2100-3 unless all parties agree to full range.

Values outside nominal black and nominal peak can exist within the permitted video data range.

This distinction matters because the number of mathematical codes is not the number of nominal visible steps from black to white in every representation.

SDR television has a similar distinction. BT.709 defines 8-bit and 10-bit television representations with nominal narrow-range values rather than assuming full range for every signal. ^[4]

The signal may not be RGB

Another simplification says that every video pixel contains one red, one green, and one blue code.

Some production and interface paths do use RGB.

Many delivery signals use components such as:

• Y'CbCr
• ICtCp
• luma-like and color-difference values

The chroma components may also be subsampled.

Bit depth describes the precision of each encoded component. It does not, by itself, tell you whether the representation is RGB, Y'CbCr, 4:4:4, 4:2:2, or 4:2:0.

That is why bit depth and chroma subsampling need separate explanations.

How many color combinations are possible?

For a full-range three-component representation:

• 8 bits per component give 256^3, about 16.7 million combinations.
• 10 bits give 1,024^3, about 1.07 billion.
• 12 bits give 4,096^3, about 68.7 billion.

These are mathematical combinations.

They are not a count of colors that every viewer can distinguish or every television can display.

The visible result is limited by:

• signal range
• transfer function
• color primaries
• chroma subsampling
• compression
• display gamut
• display color volume
• panel drive
• processing
• viewing conditions
• observer variation

The large counts show how quickly available combinations grow.

They do not prove that each combination becomes a unique visible color.

What banding actually is

A smooth digital gradient is represented by a sequence of discrete code values.

When neighboring image regions are assigned the same value, each region reproduces one level. The next code value creates a small step.

If the steps are difficult to see, the gradient appears continuous.

If the steps become visible, they form false contours.

Banding research describes these artifacts as artificially introduced contours in smooth or low-texture image regions. ^[10][11][12]

Common examples include:

• clear skies
• sunset gradients
• fog
• soft shadows
• studio backdrops
• animation
• computer graphics
• large walls
• underwater scenes
• fades to black
• game menus and skies

The artifact is easiest to notice when there is little texture to mask the contour.

Banding and posterization

The words overlap but can describe different severity.

Banding commonly refers to visible contour lines or broad steps in a gradient.

Posterization often describes a more severe reduction of tonal levels that makes an image look stylized, painted, or cut into flat regions.

Both can involve quantization.

Posterization can also be deliberate.

The article uses banding for unwanted visible contours in an intended gradient.

Where banding can enter the chain

The visible television image may pass through:

1. camera capture
2. computer-generated imagery
3. visual effects
4. color correction
5. compositing
6. graphics
7. mastering
8. export
9. compression
10. distribution
11. source-device decoding
12. color conversion
13. HDMI or internal application processing
14. television tone mapping
15. panel processing and drive

Banding can enter at more than one stage.

A ten-bit final signal does not prove that every earlier stage used ten-bit or higher precision.

Once a gradient has been reduced to visible steps, putting those steps in a larger container does not recreate the missing intermediate values.

Why a ten-bit file can contain eight-bit-origin banding

Suppose a graphic is rendered in eight-bit precision.

The gradient contains visible steps.

The finished graphic is then placed in a ten-bit timeline and exported as a ten-bit file.

The ten-bit file can preserve the eight-bit steps exactly.

It has more available codes, but the missing source values were never created.

The same can happen when:

• an eight-bit camera source is converted to ten bits
• a low-precision effect is rendered into a higher-precision master
• an eight-bit SDR element is inserted into HDR
• a web graphic is captured from an eight-bit interface
• a game renders part of its pipeline at lower precision
• a compressed source is transcoded to a ten-bit codec

Ten-bit delivery prevents additional loss better than eight-bit delivery.

It does not prove ten-bit origin.

Why an eight-bit signal can look smooth

Eight-bit video can look smooth when several conditions help it.

These may include:

• an efficient nonlinear transfer function
• a moderate luminance range
• texture or camera noise
• film grain
• dithering
• good compression
• limited screen size
• a brighter room
• shorter viewing distance sensitivity
• a gradient that does not linger
• display processing that masks the steps

This is why eight-bit SDR was usable for decades.

It is also why no single bit-depth threshold predicts every visible result.

What the vision model supports

ITU-R BT.2246-9 explains that the visibility of tone discontinuities depends on image content and viewing conditions.

It states that actual contrast sensitivity varies with:

• screen luminance
• field of view
• spatial frequency
• other conditions ^[2]

Its Barten-model examples use stated constants, angular size, integration time, and detection assumptions.

The useful engineering goal is to keep reproduced tone steps below an appropriate visibility criterion.

The article should not turn one modeled line into a universal law.

A step below a chosen threshold is not guaranteed invisible to every observer.

A step above it is not guaranteed visible in every moving image.

Why transfer functions matter

Code values are evenly spaced as integers.

They are not evenly spaced in displayed light.

A transfer function maps the codes into luminance.

An efficient function places more codes where the system needs finer visible precision and fewer where larger physical luminance changes are less noticeable.

This is why bit depth cannot be judged without the transfer function.

Eight-bit gamma-encoded SDR and eight-bit linear-light video do not have the same visible step distribution.

Ten-bit PQ and ten-bit HLG also do not distribute luminance identically.

Why HDR commonly uses more precision

HDR extends the useful signal and display range in luminance and color volume.

It places demanding gradients in:

• near black
• diffuse white
• bright highlights
• saturated colors
• large smooth skies
• computer-generated material

A simple extension of an SDR power function would not use a practical code space efficiently across the full PQ range.

SMPTE ST 2084 defines PQ as a perceptually based absolute-luminance EOTF. Its informative introduction says it creates an efficient mapping from digital values containing as few as ten bits to a range of 0 to 10,000 cd/m². ^[3]

BT.2100-3 defines both 10-bit and 12-bit integer representation for PQ and HLG. ^[1]

The standard does not say that every HDR file must be ten-bit or that ten bits eliminate every visible contour.

What HDR10's “10” means

The Consumer Technology Association's HDR10 Media Profile defines:

• SMPTE ST 2084 PQ
• 10-bit depth
• BT.2020 primaries
• defined static metadata and content-level information ^[6]

The 10 in HDR10 identifies the ten-bit profile.

It does not mean:

• the content peak is 1,000 cd/m²
• every display panel is native ten-bit
• the television processes every stage at exactly ten bits
• every source gradient began with ten-bit precision
• banding cannot occur
• ten bits are the maximum precision allowed in HDR production

HDR10+ is backward-compatible with HDR10 and adds dynamic metadata. Its current owner documentation also says the HDR10+ specification can support quantization beyond ten bits, so the HDR10 Media Profile's ten-bit limit should not be treated as a limit of the HDR10+ specification. ^[7]

PQ range and bit depth are separate

PQ's normalized EOTF range extends to 10,000 cd/m².

Bit depth determines how many digital steps are available to represent that normalized range.

HDR10+ Technologies correctly notes that the PQ luminance range does not change merely because the signal uses ten or twelve bits. ^[7]

More bits give finer quantization.

They do not change the ST 2084 endpoint.

A ten-bit PQ signal is not limited to a 1,000 cd/m² code range.

Ten-bit PQ is practical, not perfect

SMPTE ST 2084 supports efficient mapping with as few as ten bits. ^[3]

Dolby's 2014 developer paper compared ten-bit and twelve-bit PQ under its Barten-based and color-difference assumptions.

It reported that:

• twelve-bit PQ stayed below the illustrated monochrome contouring threshold
• ten-bit PQ crossed above it
• ten-bit PQ could be adequate for captured images containing noise
• graphics and noise-free images required more care ^[5]

These are attributed developer findings under the paper's conditions.

They should not be converted into either of these absolutes:

Ten-bit PQ never bands.

or:

Ten-bit PQ is inadequate for HDR.

Ten-bit distribution is a practical compromise with far more precision than eight-bit representation.

Twelve-bit offers additional margin.

Why twelve bits can matter

Twelve-bit representation provides four times as many code values per component as ten-bit.

It can help with:

• noise-free graphics
• animation
• gradients created in visual effects
• repeated transforms
• intermediate production stages
• keying
• compositing
• color-volume conversion
• aggressive grading
• archive and mezzanine masters

The final consumer signal may still be ten-bit.

Higher-precision production can reduce the chance that visible steps are created before that final quantization.

Dolby Vision and twelve bits

The statement Dolby Vision uses twelve bits needs a profile and workflow.

Dolby Vision can involve:

• higher-precision source masters
• ten-bit base layers
• enhancement information
• profile-specific codec constraints
• metadata-guided display management
• ten-bit or twelve-bit internal processing

A Dolby Vision logo does not tell the reader the native panel bit depth or every intermediate precision.

A later Dolby Vision article should cite the exact profile, container, codec, base layer, enhancement layer, and delivery specification.

The foundational advice here is simpler:

Do not use the Dolby Vision name as proof that every delivered sample is a conventional twelve-bit video stream.

What dither does

Dither intentionally changes how quantization error is distributed.

Before or during bit-depth reduction, the system adds or distributes a small noise-like pattern.

Without dither, a smooth range may round into coherent flat regions.

With suitable dither, neighboring pixels or frames can alternate among nearby codes.

The eye can integrate the fine pattern into an intermediate-looking tone, making the large contour less noticeable.

Peer-reviewed image-processing research uses probabilistic dithering to suppress contour artifacts created by color quantization. ^[10]

The trade is:

• less structured banding
• more fine noise or texture

The result can be visually preferable even though the numeric error has not disappeared.

Dither does not restore the original signal

Dither cannot recover the exact high-precision values that were discarded.

It changes the character of the error.

A dithered eight-bit image is still an eight-bit image.

It may look smoother than an undithered eight-bit image.

That is not the same as carrying the original ten-bit values.

Spatial and temporal dithering

Dither can be distributed:

• across neighboring pixels
• across frames
• through noise-shaped patterns
• through combinations of spatial and temporal variation

A static spatial pattern can appear as fine texture.

Temporal variation can create intermediate average output over time.

The visible result depends on:

• pattern
• amplitude
• frame rate
• panel response
• motion
• viewing distance
• content
• observer sensitivity

A manufacturer term such as Frame Rate Control may refer to temporal techniques, but implementation details and effectiveness are model-specific.

Compression can create or expose banding

Video compression does more than preserve or discard whole bit-depth levels.

It transforms, predicts, quantizes, and reconstructs image information.

Smooth, low-texture regions can be difficult because small changes may be treated as expendable while the resulting contours remain perceptually obvious.

Banding research continues to study artifacts introduced by quantization and video encoding even in modern systems. ^[11][12]

A ten-bit codec can still show banding when:

• bitrate is too low for the material
• quantization is aggressive
• a prior banded source is encoded
• dither or grain is smoothed away
• gradient precision was lost in preprocessing
• color conversion creates new steps
• the decoder or display adds processing

Disc versus streaming comparisons need caution

A disc may use:

• higher bitrate
• a different encode
• a different codec configuration
• different grain retention
• different metadata
• a different master
• different player output
• different television mode

A stream may use adaptive bitrate and change quality during playback.

When one source bands more than another, the difference is real to the viewer.

The cause should not be assigned until the masters, encodes, devices, and settings are compared.

Signal bit depth is not panel bit depth

Several stages need separate names.

Signal bit depth

The precision carried by the source or interface.

Decode and internal processing precision

The precision used for decompression, scaling, color conversion, tone mapping, grayscale adjustment, local dimming, and lookup tables.

Internal processing may use more precision than the input to reduce accumulated rounding error.

Panel-driver precision

The number and form of electrical drive levels sent to the panel.

Native or effective panel behavior

The distinct light outputs the panel can produce, including spatial or temporal modulation.

A television can accept ten-bit input without proving that every later stage is native ten-bit.

A panel can use temporal techniques and produce a visually smooth result without being identical to a native ten-bit panel under every condition.

The final question is measured visible performance.

Why panel specifications can be hard to interpret

Manufacturers may publish:

• accepted signal formats
• panel color count
• processing-bit claims
• driver-bit claims
• marketing descriptions
• no native panel precision at all

These claims describe different stages.

A statement such as one billion colors may be a mathematical input or processing claim rather than an independent measurement of distinct output levels.

Treat manufacturer language as documentation of the claimed feature.

Use test patterns and independent measurements for visible performance.

Can the television create banding?

Yes.

A clean source can be damaged by:

• low-precision processing
• a coarse lookup table
• poor tone mapping
• a grayscale control with large steps
• local contrast processing
• noise reduction
• edge-preserving smoothing
• color-volume mapping
• range conversion
• near-black handling
• panel drive or temporal artifacts

The same television may behave differently in:

• SDR
• HDR10
• HLG
• game mode
• PC mode
• internal applications
• HDMI inputs
• different refresh rates

Do not generalize one mode to the entire product.

Can debanding help?

Some televisions, players, and software include:

• smooth gradation
• contour reduction
• debanding
• noise reduction
• low-bitrate smoothing

These controls estimate where a contour is unwanted and smooth or dither the region.

They can improve a damaged sky or compressed stream.

They can also remove:

• real texture
• film grain
• subtle shading
• fine animation detail
• intentional posterization

Debanding is a repair or preference control.

It is not part of the original signal.

Use the lowest effective setting when the artifact is distracting, and compare it with high-quality material that contains real texture.

What an HDMI cable does

An HDMI cable transports the selected digital link signal.

It does not decode PQ, inspect the picture, and round ten-bit values to eight-bit values.

A cable that cannot reliably carry the selected rate may produce:

• loss of picture
• intermittent blanking
• sparkles or pixel errors
• audio or video dropouts
• unstable synchronization
• failure to establish the intended format

A source or system may fall back to another format because of capability negotiation or a device setting.

That fallback is performed by the source or processing device.

It is not the cable performing image conversion.

Why bandwidth still matters

Required link bandwidth depends on the complete format:

• resolution
• frame rate
• bit depth
• RGB or Y'CbCr
• chroma sampling
• timing
• compression such as DSC where supported
• audio and auxiliary data
• interface mode

A higher bit depth increases the data required when the other variables remain constant.

The system may trade among:

• frame rate
• chroma sampling
• bit depth
• compression
• resolution

This is why forcing the highest numerical bit-depth setting is not always the best overall signal.

Certified HDMI cable categories

HDMI Licensing Administrator states that:

• Premium High Speed HDMI cables are certified for the full 18 Gbit/s bandwidth used by feature-rich 4K60, HDR, and BT.2020 applications. ^[8]
• Ultra High Speed HDMI cables support system configurations up to 48 Gbit/s. ^[9]
• The HDMI 2.2 overview introduces Ultra96 HDMI cables for configurations up to 96 Gbit/s. ^[9]

Use the category required by the actual source, display, resolution, refresh rate, and format.

A 24-frame-per-second movie source does not automatically need the cable category required for 4K at 120 or 240 frames per second.

Certification verifies cable compliance to the stated program.

It does not prove that an AVR, soundbar, switch, source, and television all support the same format.

What enhanced HDMI input modes really do

Television manufacturers use names such as:

• Input Signal Plus
• HDMI Enhanced Format
• HDMI Deep Color
• Ultra HD Deep Color
• Enhanced 4K
• Enhanced 8K
• HDMI UHD Color

The names and behavior vary.

An enhanced mode may enable combinations involving:

• higher link rate
• higher frame rate
• higher bit depth
• wider colorimetry
• HDR signaling
• 4:4:4 or RGB formats
• FRL rather than older TMDS operation
• gaming features

It may also expose compatibility problems with older source or switching equipment.

The manual for the exact model, input, region, and firmware is the appropriate source.

Should enhanced mode always be enabled?

Enable the mode required for an HDR or high-bandwidth source when the manufacturer documents that requirement.

Do not assume:

• every input supports the same mode
• every source needs the mode
• the mode affects only bit depth
• the setting has the same name on every television
• turning it on repairs source banding
• turning it off always forces eight-bit output

When the chain works correctly in the required mode, leave it enabled for that source.

When it causes failure, identify the unsupported device, cable, input, or format rather than treating reduced bit depth as the only possible explanation.

Do not blindly select the highest bit-depth number

A source menu may offer:

• 8 bit
• 10 bit
• 12 bit
• Auto
• Deep Color
• output-format combinations

A twelve-bit output setting does not create twelve-bit source detail from ten-bit content.

It may cause the source to repackage or dither the signal.

At a fixed bandwidth, twelve-bit output may require a different chroma format or refresh-rate compromise.

The best goal is:

Preserve the intended source signal through a format that the complete chain supports reliably.

Auto is often a sensible starting point.

Verify it when the device provides trustworthy signal information or when test patterns reveal a problem.

A practical troubleshooting method

Step 1: Identify the content

Ask:

• Is it SDR, HDR10, HDR10+, HLG, or another format?
• Is the source known to be ten-bit?
• Is the banding visible in the original application on another display?
• Does it occur in one title or many?

One banded stream does not prove a system-wide eight-bit problem.

Step 2: Compare another source

Compare:

• internal app versus external player
• disc versus stream
• one streaming service versus another
• game console versus built-in test pattern
• downloaded test clip versus live content

Change one variable at a time.

Step 3: Check the signal range

A full-range and narrow-range mismatch changes black and contrast.

It can also make near-black steps more obvious.

Do not use bit depth to repair a range error.

Step 4: Confirm HDR mode and colorimetry

Make sure the display enters the intended HDR mode and receives the expected BT.2020-related signaling where applicable.

A transfer-function or colorimetry mismatch can look worse than ordinary quantization banding.

Step 5: Verify enhanced input mode

Follow the television manual for the exact input.

Enable the required enhanced format for the HDR or high-frame-rate source.

Step 6: Check the intermediate devices

An AVR, soundbar, switch, splitter, capture device, or extender can limit the link.

Connect the source directly to the television as a diagnostic test when practical.

Step 7: Use a certified cable appropriate to the format

Select the official cable category required for the actual bandwidth.

Replace a suspect cable when the symptoms include instability, dropouts, sparkles, or failure to establish the intended format.

Do not expect a new cable to repair banding embedded in the content.

Step 8: Check the source output

Use an output format that preserves the content and is supported by the complete chain.

Avoid unnecessary conversion.

Do not assume that a manually forced twelve-bit format is better than correct ten-bit output.

Step 9: Disable processing temporarily

For diagnosis, turn off:

• smooth gradation
• noise reduction
• dynamic contrast
• aggressive tone mapping
• edge enhancement
• adaptive color
• AI picture processing

Then re-enable one control at a time.

Step 10: Use a known gradient pattern

A useful pattern should identify:

• bit depth
• transfer function
• range
• colorimetry
• expected display mode
• source file integrity

A screenshot of a pattern can itself be reduced or color-managed incorrectly, so use the original test source rather than an image copied through a web browser.

What a gradient pattern can reveal

A clean test gradient can help separate:

• source banding
• decoder output
• display processing
• panel behavior
• range mismatch
• a mode-specific problem

It cannot prove the native bit depth of every internal stage by itself.

Temporal dithering can make a still photograph of the screen misleading.

Use:

• direct measurement
• high-frame-rate camera analysis when appropriate
• repeated patterns
• multiple stimulus levels
• mode comparisons
• manufacturer or panel documentation

What calibration can verify

Measurement and controlled testing can evaluate:

• signal range
• grayscale continuity
• near-black steps
• EOTF tracking
• clipping
• tone-mapping roll-off
• color conversion
• mode differences
• processing-induced contours
• panel repeatability
• dithering behavior

Record:

• exact model and screen size
• firmware
• input
• source
• file
• format
• bit depth
• range
• chroma sampling
• refresh rate
• picture mode
• meter
• pattern generator
• room condition
• date

Without these conditions, one gradient photograph is weak evidence.

What calibration cannot restore

Calibration cannot restore:

• source levels removed during quantization
• detail lost to compression
• gradients damaged by an earlier eight-bit render
• dither removed upstream
• clipped values
• a missing color component
• content that was already posterized

A television repair control can estimate smoother transitions.

That estimate is not the original data.

Calibration can reduce display-created error within the television's controls and physical capabilities.

Where this leaves us

Eight bits provide 256 mathematical values per component.

Ten bits provide 1,024.

Twelve bits provide 4,096.

Television video can use narrow or full-range coding, so those counts are not always the nominal black-to-white intervals.

More bit depth reduces quantization-step size.

It does not guarantee a smooth picture.

Banding can originate in capture, rendering, grading, compression, conversion, playback, television processing, or panel drive.

BT.2100 supports ten-bit and twelve-bit HDR representation.

HDR10 is a ten-bit media profile.

PQ uses the available code values efficiently, but ten-bit PQ is not guaranteed to be contour-free in every image and workflow.

Dither can trade structured contours for finer noise.

It does not restore lost source precision.

A ten-bit input does not prove a native ten-bit panel.

A weak HDMI cable does not normally convert pixels to eight bits, although the complete chain may fail or select a lower-bandwidth format.

The useful question is not only:

Is this signal ten-bit?

It is:

Where in the complete path was the gradient first reduced, compressed, remapped, or displayed as visible steps?