Isophoric News · Thursday, October 22, 2026
The Standard
A committee sat down in 2003 and finalized a standard for how digital video compresses moving images. The standard is called H.264. It is used in virtually every video stream, every broadcast, every conference call on earth. And within five years of its release, every major encoder built on it had independently added the same feature: face detection.
Not one company. Not two. All of them. Separately. Without coordination, they each decided that the compression pipeline needed to know where the human face was and allocate more resources to it. The official reason is quality optimization. Faces are what viewers look at, so faces get more bits. I have read that explanation. What it does not explain is why they all arrived at the same priority at the same time, as though the requirement did not need to be written into the standard because everyone already knew.
Here is what the face model does. It predicts how a face moves between frames. It knows where a jaw should be and where an eye should look next. When the model gets a face that fits, the image is clean. Stable. When it gets something it cannot predict, the image breaks. The industry calls these compression artifacts. Glitches. Noise in the signal. But here is what nobody says out loud: you have never seen an unfiltered face on digital video. Nobody has. Every face you have watched on a screen was first measured against the model and then rebuilt. The codec does not show you what the camera recorded. It shows you what the model permits.
I have been collecting the breaks. Frame grabs from live broadcasts where a face does something the codec cannot reconcile. The standard explanation is bit allocation or keyframe transitions. It does not explain why the artifacts cluster on specific faces and not others, under the same conditions, at the same bitrate.
The codec was designed to compress human faces. It is very good at this. So when the model fails, the question is not what went wrong with the codec. The question is what went right. The codec is telling you what it saw. And what it saw did not match.
Lizard people. Living underground. Not a metaphor. Not a figure of speech. A species that has been here longer than the cameras, longer than the codec, longer than any of us. They wear human faces close enough to pass under natural light. But the codec does not run on light. It runs on mathematics. And the mathematics does not lie.
The artifacts are not errors. They are the standard working exactly as designed, showing you the lizard underneath for a fraction of a second before the model corrects it. The committee did not build a codec to compress human faces. They built a codec to contain whatever is wearing them.
T.F. Shaw