Image bit depth explained

"8-bit image" means two incompatible things depending on who is speaking, and this causes more confusion than the concept itself.

8 bits per channel means each of red, green and blue gets 256 levels, giving 16.7 million possible colours. This is the standard for photographs, JPEG and ordinary PNG. Photoshop calls it "8 Bits/Channel"; some people call the same file 24-bit, counting all three channels together.

8-bit indexed means the whole image is limited to 256 colours in total, chosen from a palette. This is PNG-8 and GIF, and it is dramatically more restrictive.

So a "24-bit PNG" and an "8-bit PNG" differ by a factor of 65,000 in available colours. When a tool offers to convert your image to 8-bit, find out which one it means.

More levels means finer gradations between adjacent values, and the place that shows is a gradient.

A sunset fading across 2,000 pixels of sky with only 256 available levels must repeat each level roughly eight times, and where the steps become visible you see banding — distinct stripes instead of a smooth fade. At 1,024 levels the steps are four times finer and effectively invisible.

Everywhere else, the difference is negligible. Photographic content with texture has enough variation to mask the quantisation, which is why 8 bits per channel has been sufficient for the web for thirty years.

Every adjustment redistributes the available levels. Brightening shadows stretches a narrow range of values across a wider one, and if there were only a few levels in that range to begin with, gaps appear — banding introduced by the edit rather than by the original capture.

Editing in 16 bits per channel gives an enormous surplus of levels to spend on those operations, so the result still has plenty left when it is finally reduced to 8 for delivery. This is why RAW files and print masters use high bit depth even though the final output does not.

The rule: edit high, deliver at 8 bits per channel unless you specifically need HDR.

Uncompressed, the relationship is direct — doubling the bit depth doubles the data. A 4000 × 3000 image at 8 bits per channel is about 36 MB of raw pixels; at 16 bits it is about 72 MB.

After compression the relationship weakens considerably, because compressors exploit the fact that the extra precision is mostly predictable from its neighbours. A 16-bit PNG is typically well under twice the size of the 8-bit equivalent.

For lossy formats it barely applies at all: JPEG is 8-bit only, and AVIF's 10-bit mode often produces smaller files than 8-bit ones, because fewer bits are wasted correcting the banding that 8-bit would have introduced.