Video Compression Explained: Codecs, Containers, Bitrate & Keyframes

The Miracle of Video Compression

An uncompressed 4K video (3840x2160 pixels) at 60 frames per second with 10-bit color depth generates approximately **11.9 Gigabits of raw data every single second**. A standard two-hour feature film in raw format would require more than **10 Terabytes** of storage.

Yet on YouTube, Netflix, or your local phone camera, that same two-hour film fits comfortably inside a 2 GB to 6 GB file. This 500:1 compression ratio is not magic—it is the result of four decades of perceptual mathematics and spatial-temporal redundancy reduction.

Containers vs. Codecs: The Essential Distinction

One of the most frequent misconceptions among digital creators is confusing a **container** with a **codec**:

• **The Container (.mp4, .mkv, .webm, .mov)**: Think of the container as an envelope or box. It bundles the video stream, audio stream, subtitles, chapter markers, and color space metadata into a single synchronized file.
• **The Codec (H.264/AVC, H.265/HEVC, AV1, VP9, ProRes)**: The codec is the algorithm used to mathematically compress and decompress the pixels themselves.

For example, an `.mp4` file can house an older H.264 video track, a modern AV1 video track, or even a HEVC 10-bit HDR stream.

The Modern Codec Landscape Compared

How Temporal Compression Works: I, P, and B Frames

Raw video possesses two kinds of redundancy: 1. **Spatial Redundancy (Intra-frame)**: Pixels next to each other in the same frame (like a blue sky or flat wall) share nearly identical colors. 2. **Temporal Redundancy (Inter-frame)**: From one frame to the next (1/60th of a second), 95% of the scene remains unchanged; only a person's hand or mouth moves.

Codecs exploit temporal redundancy through a structure known as a **GOP (Group of Pictures)**, built from three frame types:

• **I-Frames (Keyframes)**: Completely self-contained images. They compress like standard JPEG photos without referencing any other frame. Scrubbing video player sliders jumps directly to the nearest I-Frame.
• **P-Frames (Predicted Frames)**: Store only what has changed relative to the previous frame using motion vectors.
• **B-Frames (Bi-directional Frames)**: Look both backward to previous frames and forward to upcoming frames to calculate motion interpolations, achieving maximum compression.

Rate Control Methods: CRF vs. CBR vs. VBR

How should your encoder allocate bits?

1. Constant Rate Factor (CRF) — Best for Archiving & Offline Encoding Instead of targeting a specific file size, CRF targets a **constant perceptual quality**. In complex, fast-action scenes with splashing water, the encoder automatically boosts the bitrate. In static talking-head scenes, it reduces the bitrate to near zero. - In H.264: CRF 18 is visually lossless; CRF 23 is default; CRF 28 is compact web quality. - In AV1: CRF 28–32 achieves extraordinary quality with minuscule file sizes.

2. Two-Pass Variable Bitrate (2-Pass VBR) — Best for Fixed Target File Sizes The encoder analyzes the entire video during pass 1, identifies where the most complex visual action occurs, and allocates the bit budget dynamically during pass 2.

3. Constant Bitrate (CBR) — Best for Live Streaming (Twitch, YouTube Live) Pushes a flat data rate (e.g. exactly 6,000 kbps) regardless of whether the screen is black or displaying complex 3D explosions. Mandatory for RTMP live streaming to prevent network buffer underruns.