Composition · 2026-08-13

One fully-described element per second

Scene density has an inverse relationship with quality: the more you specify, the more the model averages. The breakpoint is roughly one named visual element per second of generation — ten seconds, ten elements, maximum. Feel the urge to add more? Compress what's already there instead.

"Old women laughing" beats a paragraph about their attire, joy, and raised hands.