Bit Depth in Photography: 8-Bit, 16-Bit and Raw

Bit depth is the number of bits a digital image uses to store each color channel of each pixel, and it sets how many separate tones the file can hold between black and white. An 8-bit channel has 256 levels, a 14-bit raw file has 16,384, and a 16-bit file has 65,536. It matters most while you edit: more levels mean finer steps, so a sky or a shadow can be pushed a long way and still look smooth instead of breaking into stripes.

A standard JPEG is 8 bits per channel. Most camera raw files record 12 or 14 bits. Editing in 16-bit mode carries that extra precision through every adjustment, and you drop to 8 bits only when you export the finished photo.

Levels, not range: the staircase model

A bit is a single binary digit, 0 or 1. Each extra bit doubles the number of values it can store, so the number of levels in a channel is 2 raised to the bit depth. That is the whole formula: levels = 2ⁿ, where n is the bit depth. One bit gives 2 levels, pure black and pure white. Four bits give 16. Eight bits give 256, numbered 0 to 255.

Picture a staircase between two floors. The height between the floors is fixed: black at the bottom, the brightest white the file can show at the top. Bit depth does not move either floor. It only sets how many steps the staircase has. A file with more bits does not have deeper blacks, brighter whites or more saturated colors. It has smaller steps between them.

A color photo has three channels, red, green and blue, each with its own levels. At 8 bits per channel that is 256 × 256 × 256 = 16,777,216 possible colors. How color counts grow with bits, and what screens can show, is covered on color depth.

Bits per channel versus bits per pixel

The same file can be described with two different numbers. Bits per channel is the number editing software uses, and the one on this page: 8-bit, 16-bit. Bits per pixel adds the channels together. So 8 bits per channel RGB is also called 24-bit color (8 × 3), and 16 bits per channel is 48-bit color. A “32-bit” image is usually 8 bits each for red, green and blue plus an 8-bit transparency channel, and is no smoother than an ordinary 8-bit file. A grayscale image has one channel, so 8-bit grayscale means 256 shades of gray. Whenever a setting or a spec gives a number, check which of the two it means.

Bit depth by file type and stage

• 8: Levels per channel: 256. Where you meet it: JPEG, most web images, most screens, most print files.
• 10: Levels per channel: 1,024. Where you meet it: 10-bit HEIF photos, some video, HDR and wide-gamut screens.
• 12: Levels per channel: 4,096. Where you meet it: Many raw files, and raw files from some cameras in fast burst modes.
• 14: Levels per channel: 16,384. Where you meet it: Raw files from most interchangeable-lens cameras.
• 16: Levels per channel: 65,536. Where you meet it: 16-bit TIFF, PSD and PNG files; a few raw formats.
• 32 (floating point): Levels per channel: Not a fixed count. Where you meet it: HDR merges, panorama and compositing work.

• Raw: 12 or 14 bits, sometimes 16, set by the camera’s analog-to-digital converter. What a raw file holds explains the rest of its contents.
• JPEG: 8 bits per channel. The standard allows a 12-bit version, but cameras, browsers and editors almost never use it, so in practice every JPEG you meet is 8-bit.
• HEIF: usually 8 or 10 bits per channel. Phones use the 10-bit mode for HDR photos. See HEIF and HEIC.
• TIFF: 8 or 16 bits per channel, or 32-bit floating point. The usual container for a 16-bit master.
• PNG: 8 or 16 bits per channel, lossless.
• PSD: 8, 16 or 32 bits per channel, with layers.

Two details are worth knowing. A “16-bit” document in some pixel editors actually stores values from 0 to 32,768, which is 15 bits plus one value; that is still about eight times finer than a 12-bit raw and makes no visible difference. And 32-bit floating point is a different kind of number altogether: instead of a fixed count of steps between black and white, values can run past white, so an HDR merge can hold highlights brighter than the screen’s white until you decide how to fit them in. The image file formats guide compares the formats on everything besides bit depth.

Why 8 bits is enough to look at but raw needs 12 or 14

This is the question most explanations skip, and it is the key to the whole subject. The answer is in how the levels are spread out.

A camera sensor is linear: twice the light gives twice the number. Your eyes are not. You see each halving of light, each stop, as roughly the same step darker. Store light linearly and half of all the levels describe the single brightest stop, a quarter describe the next stop, and the darkest stops get almost nothing.

A JPEG avoids that by storing its 256 codes on a gamma curve, which spaces them much closer to the way you see. The chart below counts how many of the 256 codes land in each stop below white, first for linear storage and then for the sRGB curve a JPEG uses.

That is why a finished 8-bit JPEG can show a smooth sky and still have tones left for the shadows. On a large, smooth, dark gradient a careful eye can sometimes just pick out 8-bit steps on a good screen; at 10 bits they are generally invisible.

Raw data is stored linearly, so it needs far more bits to give the shadows enough levels. Here is how many levels each stop gets in a linear raw file:

Each figure is the file’s total levels divided by 2 for every stop down: 16,384 ÷ 2 = 8,192 for the first stop of a 14-bit file, 16,384 ÷ 256 = 64 for the eighth. When a raw file is converted, the software applies the tone curve, white balance and color, then rounds the result. Converted to an 8-bit JPEG, the extra levels are gone for good, which is one of the clearest reasons to keep the raw file.

Some cameras do not store raw data in plain linear form. They compress it, sometimes with a curve that drops levels in the highlights where there are far more than needed. In those files the per-stop counts above are only a guide.

Bit depth is not dynamic range

Dynamic range is the span from the darkest to the brightest tone a sensor can record with usable detail, measured in stops. Bit depth is how finely the file divides whatever span it holds. They are linked by one ceiling: a linear file of n bits can hold at most about n stops, because the smallest step above zero is 1 level and the brightest value is 2 to the power n. A 12-bit linear raw has room for about 12 stops, a 14-bit one about 14.

In practice the sensor’s noise, not the file, sets the real limit, and many sensors deliver fewer usable stops than their bit depth. Adding bits to a camera does not add dynamic range. It records the range the sensor already has in finer steps.

Wondering how many stops your scene actually spans, and what to do when it is more than the sensor can hold? Dynamic range measures the scene against the sensor and shows the ways to fit it.

Noise, not levels, is the usual limit in raw shadows

The table above makes it tempting to blame shadow problems on a shortage of levels. Usually that is wrong. Real sensors produce noise, random variation from the light itself and from the electronics, and in a raw file that noise is normally larger than one level step. Noise scatters neighboring pixels across several levels, which hides the steps the same way a printer’s dots hide the gaps between ink colors. This scattering is called dithering, and it is why raw files almost never show banding from their own bit depth.

So when lifted shadows look rough, the cause is noise, and the fix is more light: a brighter exposure, or exposing to the right without clipping the highlights. That helps by raising the signal above the noise, not by reaching more levels.

Should you shoot 12-bit or 14-bit raw?

Many cameras let you choose. 14-bit records finer steps underneath the noise, so it can make a small difference when you lift deep shadows by several stops on a low-ISO exposure of a high-contrast scene. 12-bit files are smaller, and on some cameras the fastest burst rates and the electronic shutter record only 12 bits.

As ISO rises, the difference disappears, because noise grows to many level steps wide and dithers away anything the extra two bits would add. A sensible default: 14-bit for landscapes, architecture and anything you expect to edit hard; 12-bit is fine for sports, events and high-ISO work, where speed and card space matter more.

Editing in 8-bit or 16-bit

Every edit is arithmetic, and the result of each step is rounded back to the nearest level the file has. Stretch the tones, as you do when you brighten shadows or add contrast, and the levels are pulled apart: where an 8-bit sky had values 140, 141, 142, a strong curve may turn them into 140, 143, 146, with nothing in between. Each jump is a visible edge. Stack several adjustments and the rounding errors add up. With 65,536 levels the same edits leave steps far too small to see.

• Edit in 16-bit whenever the file came from raw and you plan more than light adjustments. Raw converters work at high precision internally; the choice matters when you send the file on to a pixel editor, so choose 16-bit there.
• Convert an 8-bit JPEG to 16-bit before heavy work. It cannot bring back tones the JPEG never had, but it stops the new edits from adding rounding errors of their own, so banding grows more slowly from that point on.
• Export in 8-bit for the web, social media, email and most print labs. Screens and browsers display 8 bits, and an 8-bit export of a 16-bit edit looks smooth because the stretching was done with full precision. Many editors also add a faint dither noise when they convert to 8 bits, which hides any remaining steps.
• Keep a 16-bit master (TIFF or PSD) of important work so you can re-edit without starting again from raw.
• Pair wide color spaces with 16 bits. Bit depth and gamut are separate: the color space sets how far the colors reach, bit depth sets how many steps fill that reach. ProPhoto RGB spreads 256 levels over a much wider range than sRGB, so each step is bigger. See color space.

The cost is file size. An uncompressed 24-megapixel RGB image takes about 72 MB at 8 bits per channel (24 million pixels × 3 channels × 1 byte) and about 144 MB at 16 bits (2 bytes per channel). Image resolution and file sizes shows how to work out others.

Banding: what too few levels look like

When an 8-bit file is stretched too far, a smooth gradient, a clear sky, fog, a studio backdrop or a sunset, turns into visible stripes of flat color. This is banding, also called posterization: steps of flat tone where there should be a smooth blend. In a histogram it shows as a comb: thin spikes with empty gaps between them, because some levels no longer have any pixels.

Clean, low-ISO files of smooth subjects band soonest, because they carry so little noise to dither the steps. Banding is most likely when you:

• Brighten shadows or pull down highlights a long way on a JPEG.
• Apply strong curves, contrast or color grading in 8-bit mode.
• Work in a wide color space such as ProPhoto RGB at 8 bits.
• Stack many adjustments, each rounding its result.

The most common route to banding in a finished photo is the export: an image edited hard, then saved as an 8-bit file, where the last rounding lands on a gradient that has already been stretched. Editing in 16-bit and converting to 8 bits only at the end, with the editor’s dither on, prevents most of it.

JPEG banding and other look-alikes

Not every stripe is a bit-depth problem. JPEG compression works on small blocks of pixels and smooths away fine variation inside each one, so at low quality settings a gentle sky breaks into blocky steps even when the levels were all there. Saving at a high quality setting, and not re-saving the same JPEG again and again, avoids it. Some screens and screen settings also show banding in gradients that are smooth in the file. Check the photo at 100 percent on a second screen, and look at the histogram, before you rework it.

Seeing dark bands across a photo taken under LED or fluorescent light? That is flicker from the light source, not bit depth: mechanical vs electronic shutter explains when each shutter shows it and how to avoid it.

Preventing and fixing banding

• Prevent it at capture. Shoot raw, and get the exposure close in camera so the sky does not need a huge correction later. A graduated ND filter balances a bright sky before the shot instead of in the edit.
• Edit in 16-bit and drop to 8 bits only for the final export.
• Add a little grain or noise to a gradient that has already banded. The random variation breaks up the edges between steps, the same way dithering does; a fine amount at 100 percent is usually invisible at normal viewing size.
• Re-edit from the source if you can. Going back to the raw file or a 16-bit master and redoing the adjustment is better than any repair to the banded copy.
• Blur only as a last resort. A gentle blur confined to the sky can smooth the steps, but it softens anything else it touches.

Screens, printers and bit depth

Most computer screens display 8 bits per channel. Some wide-gamut and HDR screens show 10 bits, either natively or by flickering between neighboring 8-bit levels. A full 10-bit chain (screen, graphics card and software all supporting it) shows gradients more smoothly while you edit, but it does not change the file. A screen that bands on its own can make you chase a problem that is not in the photo, which is one reason calibrating the monitor is worth doing. Most printers and labs work from 8-bit files and print them smoothly, provided the editing was done at higher precision.

Common mistakes with bit depth

• Thinking more bits means more dynamic range. A 16-bit file holds no more highlight or shadow detail than the sensor captured. It holds the same detail in finer steps.
• Thinking more bits means richer or more saturated color. How far colors reach is set by the color space. Bit depth only sets how finely that range is divided.
• Reading 24-bit as better than 8-bit. 24-bit color is 8 bits per channel, counted per pixel. A “32-bit” PNG is the same thing with transparency.
• Sending a raw file to a pixel editor as 8-bit. Choosing 8-bit output throws away most of the levels before the real editing starts.
• Saving the only edited copy as a JPEG. Every later edit starts from 8 bits plus compression. Keep a 16-bit master of anything that matters.
• Expecting 16-bit to repair a JPEG. Converting up limits further damage; it cannot restore tones the JPEG never stored.
• Exporting 16-bit files for the web. Browsers and social sites display 8 bits; a 16-bit upload is larger and is converted anyway.
• Blaming bit depth for every stripe. Compression, the screen and noisy shadows all cause problems that look similar.

Try this: break a sky on purpose

Take a raw photo of a clear sky, a patch of fog or a plain wall with soft light falling across it. If your phone can save raw (often as DNG), use it. From your raw converter, export one copy as an 8-bit JPEG and one as a 16-bit TIFF, and open both in an editor that can work in 16-bit. On each, apply the same extreme Curves adjustment: lift the shadows hard and add strong contrast.

Judge the result in three places. Look at the histogram: the 8-bit copy should show a comb of spikes and gaps, the 16-bit copy a solid shape. Look at the gradient at 100 percent: stripes in the 8-bit copy, smooth tone in the 16-bit one. Then export the 16-bit copy as an 8-bit JPEG and check it again; it should stay smooth, which shows that precision during editing is what counts. If you see no difference at all, push harder, or pick a cleaner, low-ISO gradient: noise may be dithering your steps.

Questions about bit depth

What does bit depth mean in a photo? It is the number of bits used to store each color channel of each pixel, and it sets how many tonal levels the image can hold. That is 256 per channel at 8 bits, 16,384 in a 14-bit raw and 65,536 at 16 bits.

Is 8-bit or 16-bit better for photos? Edit in 16-bit, especially from raw, then export 8-bit copies for sharing and most printing. The 16-bit edit keeps gradients smooth through heavy adjustments, and the 8-bit export keeps files small and works everywhere.

What is the bit depth of a JPEG? 8 bits per channel, which is 24 bits per pixel for the three color channels together. That gives 256 levels per channel and about 16.7 million possible colors.

Can I see the difference between 12-bit and 14-bit raw? Rarely. It can show when you lift deep shadows several stops on a clean, low-ISO file, and it vanishes at higher ISO, where noise is coarser than the extra steps.

Is an HDR photo the same as a high bit depth photo? No. HDR describes a scene or a display with a wide range of brightness. Holding that range smoothly takes more bits, which is why many HDR photos are stored as 10-bit files and HDR merges use 32-bit floating point, but a high bit depth file of an ordinary scene is not HDR.

How do I check or change the bit depth of an image? To check, look at the file’s information panel in your operating system or editor, and note which number it gives: many systems report bits per pixel, so an ordinary JPEG shows as 24 and a PNG with transparency as 32, both of which are 8 bits per channel. To change it, use the image mode setting in a pixel editor to switch between 8 and 16 bits per channel, or choose the bit depth in a raw converter’s export or hand-off settings. Going down throws levels away for good; going up adds no tones the file never had.

What does bit depth mean for video? The same thing: levels per channel. Most video is 8-bit, and 10-bit video gives four times as many levels, which matters when footage is shot flat (log) and graded heavily afterward, for the same reason 16-bit matters when editing stills.

Deciding whether to shoot raw at all? Raw vs JPEG lays out what each format keeps and when the smaller file is the better choice.

Want to check you can tell a raw file’s strengths from a JPEG’s? The raw vs JPEG quiz tests what you know about file formats and resolution, with an explanation for every answer.