Step-Video-T2V

class B live

30B open-weights text-to-video model (~8s) with a deep-compression VAE; MIT-licensed, self-host plus free online demo.

Layer 1 — factual axes

API & price	self-host only	official	source ↗ 2026-07-04
Max output	204 frames (~8s) · 544×992	official	source ↗ 2026-07-04
Native audio	no	official	source ↗ 2026-07-04
Weights	open · MIT	official	source ↗ 2026-07-04
Free tier	open self-host (free demo at yuewen.cn)	official	source ↗ 2026-07-04

No independent measure yet. We publish a quality figure only when a named external benchmark ranks this model — we don’t invent one.

30B open-weights text-to-video model (~8s) with a deep-compression VAE; MIT-licensed, self-host plus free online demo.