Alaya-EVOKE: From Linear-Scaling Supervision to Endless World
2026-08-13 • Computer Vision and Pattern Recognition
Computer Vision and Pattern Recognition
AI summaryⓘ
The authors developed Evoke, a system for generating interactive world models that can remember long histories without slowing down or losing important details. They separate the world’s state into an external memory bank and redesign the teacher model to handle long sequences efficiently by using clever attention methods. This lets Evoke generate new views quickly while keeping the whole story consistent over time. Their approach improves long-term memory and control in the model and achieves strong performance on benchmark tests.
world modelpersistent memorydenoiser contextkey-value cacheattention mechanismlong-horizon generationexternal memoryteacher-student modelself-forced rolloutsclassifier-free guidance
Authors
Yuanyang Yin, Gongxuan Wang, Yifan Zhan, Chuanhao Li, Kaipeng Zhang, Feng Zhao
Abstract
Interactive world models must support persistent memory, responsive interaction, and long-horizon generation, yet these requirements place conflicting demands on the model. Maintaining history in the denoiser context or key-value cache incurs growing cost, forcing a trade-off between session length and retained memory, while low-latency interaction relies on few-step generation whose capabilities are bounded by its teacher. Evoke addresses both limitations by externalizing persistent world state and redesigning the teacher for long-horizon interactive generation. Scene geometry is maintained in an external, camera-indexed world state bank, from which only view-relevant information is retrieved, keeping the denoiser context bounded as the session grows. Rather than treating the teacher as a fixed generator, we design it for long-horizon supervision: its sparse attention combines chunk-wise grouping, retrieval of selected distant frames, and a linear-attention global state, yielding linear growth in memory and compute while enabling supervision over long horizons. Such supervision exposes content drift that stays locally plausible within short windows, while per-chunk conditioning enables prompt changes and event control throughout the sequence. A 30-second distribution-matching objective, applied under self-forced rollouts, transfers both capabilities to a three-step student that uses no classifier-free guidance, improving resistance to long-term drift while preserving responsive conditioning. With bounded context and recurrent external memory, Evoke supports open-ended, continuously evolving generation; on a single H200 at $384\times 640$, each $1.5\,\mathrm{s}$ chunk is generated in $2.11\,\mathrm{s}$. As a three-step world model, Evoke achieves state-of-the-art performance on WBench while remaining competitive on VBench-Long and VBench-2.0.