H.264 Is Magic
October 30, 2026, Noon
A 300-frame video encoded with H.264 takes 175 KB. A single frame of that same video saved as a PNG takes 1'015 KB. The video stores 300 times the data in a fifth of the space. That makes H.264 roughly 1'500 times more efficient than PNG per frame [1].
How? H.264 is the result of over 30 years of cumulative engineering with a single goal: reduce the bandwidth required to transmit full-motion video. Without it, a 50 GB Blu-ray disc would hold about two minutes of uncompressed footage. Even SSDs struggle to dump raw video from RAM to disk fast enough [1].
Lossy Compression: Shedding Weight
The first trick is throwing away what you don't need. PNG is lossless, meaning every pixel can be recovered bit for bit. H.264 is lossy. It discards detail that the human eye barely notices. The speaker grills on a MacBook Pro disappear in a close-up, but at normal viewing distance you'd never tell. The compressed image weighs 7% of the original, and that's before traditional compression even begins [1].
Inter-Frame Prediction: Not Every Frame Is New
Most video frames are nearly identical to the ones before them. A person talking on camera moves their mouth, but the background stays fixed. H.264 doesn't store each frame independently. It stores a reference frame (an I-frame) and then describes subsequent frames (P-frames and B-frames) as changes from earlier or later frames. If a block of pixels hasn't moved, the codec just says "same as before" [1].
Motion Vectors: Tracking Movement
When objects move, H.264 doesn't re-encode them. It encodes a motion vector, a mathematical description of where a block of pixels moved from and to. Instead of storing a new 16x16 block of pixel data, it stores a tiny vector pointing to the previous location plus a residual difference. This is why panning shots compress so well [1].
Transform and Quantization
After prediction, H.264 applies a discrete cosine transform (DCT) to the residual data, converting spatial information into frequency information. High-frequency detail that humans barely see gets quantized, meaning the codec rounds it off aggressively. Low-frequency data that matters is preserved carefully. This is where most of the size reduction happens [1].
Entropy Coding: The Final Squeeze
Everything that remains gets passed through context-adaptive binary arithmetic coding (CABAC) or context-adaptive variable-length coding (CAVLC). These are statistical compression methods that assign fewer bits to common patterns and more bits to rare ones. It's the same principle as Huffman coding but far more sophisticated, adapting to the statistics of the actual content [1].
Why It Matters
H.264 (also known as AVC, Advanced Video Coding) was finalized in 2003 and remains the dominant video codec on the planet. YouTube, Netflix, Blu-ray, FaceTime, most web video, and every phone camera output use it. Its successor HEVC (H.265) and the open AV1 codec improve efficiency, but H.264's combination of quality, speed, and universal hardware support keeps it everywhere [2].
Every time you watch a video on your phone, a pipeline of techniques invented across three decades runs in real time on a chip smaller than your fingernail. That's not just engineering. It's magic [1].