Bit Depth, Channels, and Alpha

A digital image is a grid of pixels, and each pixel is a small set of numbers - typically red, green, blue, and often a fourth, alpha. The colorful physics and perception elsewhere in this library all collapse, at storage time, into integers like (220, 40, 40, 255). Everything practical about digital color - smoothness, transparency, file size, fidelity - comes down to three questions about those numbers: how many channels, how many bits each, and how they are interpreted.

Get those right and color "just works." Get them wrong and you meet the classic bugs: striped skies, muddy edges, posterized exports, and files ten times bigger than they need to be.

Bit depth is how many bits store each channel, which fixes how many distinct levels that channel can take: 2^bits. Eight bits give 256 levels per channel and 256³ ≈ 16.7 million RGB colors. Drop the bit depth and the steps between levels grow until they become visible as banding across a smooth gradient.

A color image is really three (or four) grayscale images stacked together - one per channel. Each channel records how much of that primary is present at each pixel. Editors expose this directly: you can view, adjust, or swap a single channel, and "grayscale" is just the channels collapsed into one luminance image.

The fourth channel, alpha, does not store color at all - it stores opacity, how much of the pixel is "there." On its own it does nothing; it tells the compositor how to blend this layer over what is behind it, using the "over" operator. Alpha is what makes transparency, anti-aliased edges, and layered UI possible.

There are two ways to store an RGBA pixel, and mixing them up is a classic source of ugly edges. With straight (unassociated) alpha, the RGB values are the true color and alpha is kept separate. With premultiplied (associated) alpha, the RGB values are already multiplied by alpha, so a 50%-opaque red is stored as half-strength red. Premultiplied is what compositors use internally because it makes filtering correct: when you scale, blur, or anti-alias an edge, premultiplied values interpolate without pulling in the color of fully transparent pixels.

Not every image stores full RGB per pixel. Indexed color (as in GIF and PNG-8) keeps a small palette of colors and stores, per pixel, just an index into it - a few bits instead of 24. The trade-off is quantization: every original color must snap to the nearest palette entry, so gradients posterize and detail is lost. Shrink the palette and watch the image collapse onto fewer and fewer colors.

How many bits you spend per channel is a budget against banding, editing headroom, and HDR. Eight bits is the web baseline and fine for final, gamma-encoded display images. But edit aggressively - big curves or exposure moves - and 8-bit steps get stretched into visible bands, which is why editing and HDR want more.

Six questions on bit depth, channels, alpha, and indexed color. Instant feedback, no scores recorded - a wrong answer comes with a short explanation.

Storage sits beneath encoding, gradients, and HDR. The numbers reflect each article's position in the editorial roadmap.