Streaming platforms shrink a movie by throwing away data your eyes and ears are least likely to miss, then spending the saved bits where you would notice a loss. Modern video codecs — the compression formats that encode and decode a picture, such as H.264, HEVC and AV1 — do this by storing only what changes between frames, by describing those changes in blocks rather than pixel by pixel, and by letting the encoder vary its effort shot by shot.
The result is not literally lossless. A two-hour film that occupies tens of gigabytes as a studio master arrives over your connection at a few gigabytes or less, and some information is gone for good. What makes it look intact is that the discarded information was chosen with human perception in mind, and that the remaining bits are allocated unevenly: a dark, grainy, fast-moving scene gets far more than a static dialogue shot.
Table of Contents
- Why most of a movie is redundant
- What gets thrown away, and why you don't see it
- Per-title and per-shot encoding
- Adaptive streaming is doing half the work
- What a viewer can actually control
- Frequently Asked Questions
Why most of a movie is redundant
A film is a sequence of still images shown roughly 24 times a second, and consecutive images are nearly identical. In a dialogue scene, the room, the lighting and the furniture persist for minutes; only faces and mouths move. Storing each frame in full would mean storing the same wall hundreds of times. Codecs exploit this with inter-frame prediction.
A keyframe stores a complete picture, and the frames after it store only motion vectors — instructions saying "this block of pixels moved slightly left" — plus a small residual for whatever the prediction got wrong. That is why a slow, steady shot compresses dramatically and a handheld chase through confetti does not. There is redundancy inside a single frame too. Smooth gradients, flat walls and large areas of similar colour can be described far more compactly than they can be listed pixel by pixel. Intra-frame prediction and frequency-domain transforms handle that part, converting blocks of pixels into coefficients where the visually important information concentrates in a few numbers.
What gets thrown away, and why you don't see it
The lossy step is quantization: the encoder rounds those transform coefficients, keeping the coarse structure of a block precisely and the finest detail approximately. Human vision is far more sensitive to broad shapes and edges than to very fine high-frequency texture, so rounding the fine detail is the cheapest possible sacrifice. Colour takes a second, larger cut. The eye has more receptors for brightness than for colour, so nearly all delivered video uses chroma subsampling — commonly written 4:2:0 — which stores full-resolution brightness but colour information at half resolution horizontally and vertically.
This alone removes a large share of the data before any other compression happens, and it is essentially invisible on normal content. The limits show up where perceptual assumptions break down. Film grain and smoke are high-frequency noise that prediction cannot anticipate and quantization flattens, which is why grainy films can look waxy or develop blotchy patches. Dark scenes expose banding, where a smooth gradient turns into visible steps, because there are few code values to work with in the shadows. Some codecs, AV1 among them, address grain specifically by stripping it before encoding and synthesizing a similar grain pattern at playback, which costs almost nothing in bitrate.
Per-title and per-shot encoding
The older approach was a fixed ladder: every title encoded at the same set of resolutions and bitrates. That wastes bits on simple content and starves complex content, because an animated short and a war film do not need the same budget to look equally good. Per-title encoding analyses each film and builds a ladder suited to it. Per-shot encoding goes further and varies the settings within the film, measuring each shot's difficulty and assigning a bitrate accordingly.
The practical effect is that the easy 80 percent of a movie subsidizes the hard 20 percent, and average file size drops while the worst moments improve. Deciding "equally good" requires a quality metric that tracks human judgement rather than raw pixel error. Perceptual metrics such as VMAF and SSIM were developed because peak signal-to-noise ratio rewards encodes that look worse to viewers. Encoding pipelines target a perceptual score, re-encode anything that falls short, and treat the metric as a proxy that still needs spot-checking by eye.
Adaptive streaming is doing half the work
Even a well-compressed film has to survive a variable connection. Adaptive bitrate streaming solves this by cutting the movie into short segments — typically a few seconds each — and encoding every segment at several quality levels. Your player measures throughput and buffer level and requests the next segment at whatever level it believes it can download in time. This is why a film often starts soft and sharpens within seconds: the player opens conservatively to avoid a stall, then climbs.
It is also why picture quality can drop mid-scene without anything buffering. The switch is the protection against buffering. Two consequences matter for anyone judging a film's presentation. First, what you saw may not be the best encode available — a congested evening, a shared connection or an overloaded Wi-Fi band can hold you below the top rung for an entire viewing. Second, the top rung itself differs by platform, by device and by subscription tier, so the same title can be a meaningfully different object on a phone app, a browser tab and a television.
What a viewer can actually control
A fair amount of visible quality loss comes from the delivery end rather than the encode, and that part is addressable: For film analysis specifically, treat a streaming image as evidence with known distortions. Grain structure, shadow detail and the texture of a dissolve are exactly the attributes compression alters most, so an argument about a cinematographer's choices built on a mid-bitrate stream is standing on uncertain ground. A disc release, where the bitrate is fixed and generous and nothing adapts to your network, is the sturdier reference when fine texture is the point.
- Use a wired connection or the 5 GHz band for the television, and check that nothing else is saturating the line during a film.
- Check the app's own quality setting; several default to a data-saving mode, especially on mobile and on cellular.
- Confirm your plan's maximum resolution. Cheaper tiers are often capped below 4K regardless of your hardware and connection.
- Prefer the native app on a television over casting from a phone or a browser, where the highest tiers are frequently unavailable for licensing or playback reasons.
- Give playback a few seconds before judging the image, since the first segments are deliberately conservative.
Frequently Asked Questions
Is 4K streaming the same quality as a 4K Blu-ray?
No. Both are 4K in resolution, but a disc carries a far higher bitrate with no adaptation to network conditions, so fine texture, grain and shadow gradients hold up better.
Why do dark scenes look worse than bright ones?
Shadows use a narrow range of code values, so quantization there produces visible banding and blotching. Dark scenes are also often grainy, which compresses poorly.
Does a newer codec mean better picture at the same bitrate?
Generally yes — newer codecs achieve similar perceptual quality at lower bitrates — but only if your device supports hardware decoding for it. Otherwise the platform falls back to an older codec.
Why does the picture get sharper a few seconds in?
The player requests a low-quality segment first to start playback without stalling, then steps up the quality ladder as it measures your available bandwidth.