Why Video Compression Looks Worse in Dark Scenes

The short answer

Dark scenes tend to look worse after compression for three connected reasons. A camera in low light amplifies its signal, which adds random noise that is expensive to encode. Shadows also contain only a small number of distinct brightness values, so any rounding by the encoder shows up as visible steps and blocks. And encoders aim to save data where they expect viewers not to notice, but dark areas are often where the damage becomes easiest to see. In a video call, all of this happens on a bitrate budget that is usually tight, so the effects are amplified. The main exception is a dark scene that is clean and well exposed, such as a properly lit subject against a dark background, which can compress reasonably well.

For readers focused on webcam audio, the key point is that this is a picture problem. Video and speech travel as separate streams with separate codecs, so a dim room does not degrade the sound itself. The sections below explain the mechanism, where the indirect links to speech are, and what to adjust.

How lossy video compression decides what to throw away

Common conferencing codecs, including H.264, VP8, VP9 and AV1, split each frame into blocks, predict each block from nearby pixels or earlier frames, and then encode only the difference. That difference is converted into frequency components and quantized, meaning it is rounded to a coarser set of values. Quantization is where most of the data savings happen and where most visible artifacts come from. At low bitrates the rounding is aggressive, so fine detail disappears and the edges between blocks can become visible as a grid pattern, often called blockiness or macroblocking.

Why darkness makes each step harder

• Sensor noise: in low light a webcam raises its gain or lengthens exposure. Gain amplifies noise along with the image, and that noise changes randomly from frame to frame. Because noise cannot be predicted from neighbouring pixels or earlier frames, the encoder either spends many bits on it or smears it away, and on a limited budget it usually does some of both.
• Few brightness levels in the shadows: standard 8-bit video uses a limited range of code values, and gamma encoding places relatively few of them in the darkest tones. A smooth dark gradient, such as a shadowed wall, may be represented by only a handful of steps. Quantization merges neighbouring steps, producing visible bands and flat patches.
• Encoder tuning: many encoders apply psychovisual or adaptive quantization that moves bits toward regions judged more important. Depending on how a given encoder is configured, flat dark areas may receive fewer bits because they are assumed to be less noticeable, which is precisely where banding and blocking then appear.
• Smearing over time: to keep noisy dark frames within budget, an encoder may lean heavily on copying earlier frames, so dim areas can look smeared or seem to crawl when something moves.
• Viewing conditions: on a bright screen in a dim room, or after the receiving device or app brightens the image, small shadow errors become far easier to see than errors of the same size in bright areas.

These factors reinforce one another. Noise uses up bits, fewer bits mean coarser quantization, and coarser quantization falls hardest on the shadow tones that had the least precision to begin with.

What this means for capturing speech

Audio quality does not depend on scene brightness. The webcam microphone, any operating system processing and the conferencing app's audio codec work independently of the image, so a dark room will not make speech muffled or noisy. If speech does sound poor, look at the microphone, its input level and the app's noise suppression settings rather than the lighting.

There are two indirect connections worth knowing about, both stated here as general principles rather than measured results. First, audio and video share the same network connection, and a noisy, hard-to-compress dark picture demands more bandwidth. How a particular app splits bandwidth between audio and video, and whether it protects audio when the link is congested, varies by application and was not established in the material available for this article. Second, listeners use lip movements to help follow speech, particularly in noisy settings or when audio is imperfect. If a dark, blocky image hides the mouth, the speaker may become harder to follow even though the audio stream has not changed. This is an inference about perception, not a test of any specific product.

Practical steps to reduce dark-scene artifacts

• Light your face from the front, for example with a lamp or window facing you rather than behind you. More light lets the camera use less gain, which reduces noise at its source.
• Avoid strong backlighting. A bright window behind you can make the camera expose for the background and leave your face in shadow.
• If your webcam software provides manual exposure or gain controls, try a lower gain setting once the lighting is adequate. Control names and availability vary by camera and software.
• Keep the scene simple. A plain, evenly lit background gives the encoder less to work on than a cluttered, dim room.
• Check your connection. A slow or congested network lowers the bitrate available for video and makes every artifact worse, in dark scenes most of all.
• Check the result as others see it. Many conferencing apps offer a self-view or test call, though a local preview may look cleaner than what remote participants actually receive, so a short recorded test or a trial call with someone else gives a more reliable picture.

Limits of this explanation

This guide covers general, well-established principles of lossy video coding and camera behaviour. No specific measurements, product specifications or app-by-app bandwidth policies were available, so it does not rank webcams, quote bitrates or claim how any particular conferencing app prioritises audio over video. To find out how your setup behaves, the most reliable approach is a direct comparison: record or test a call in dim light, then again with your face well lit, and compare both the picture and the clarity of your speech.