Alaya-EVOKE: 14B World Model Generates 120-Minute Videos on One GPU

AlayaLab, with USTC and Shanghai Innovation Institute, released EVOKE, a 14B autoregressive world model enabling 3-step CFG-free generation at 384×640@24fps. It demonstrated 120-minute continuous generation—4800 clips and 172,800 frames—on a single GPU.

2026-08-16 ~ 2026-08-17 · 2 related posts

• EVOKE 14B World Model: 3-Step CFG-Free Interactive Generation — Crazy-Repeat-2006 · 2026-08-16
• Alaya-EVOKE: a video world model that generates 120 minutes straight on a single GPU — 量子位 · 2026-08-17