Generative video models have advanced considerably in fidelity, opening the door to long-horizon generation, interactive world modeling, and evolving visual environments. Autoregressive (AR) video generation, in particular, extends visual sequences through causal rollouts. Yet a fundamental bottleneck persists: as generated sequences grow, practical models must operate under strictly bounded context windows, storage, and computational budgets.
According to a new survey published on Hugging Face, this constraint means critical historical information — such as entity identities, dynamic states, and intervention-induced causal changes — often exits the active context long before its relevance fades. The authors argue that overcoming this limitation and maintaining temporal persistence is fundamentally a memory problem.
The paper offers a systematic and comprehensive review of memory mechanisms in AR video generation. It defines memory operationally as persistent historical information maintained across outer AR steps, capable of influencing future generation even after the originating evidence is no longer locally accessible.
Building on this unified framework, the survey organizes the literature through five complementary perspectives:
- Forms — the representational carriers of history.
- Functions — the specific semantic and physical information requiring preservation.
- Operations — the lifecycle of writing, reading, updating, managing, and integrating memory.
- Learning — the optimization of memory behaviors under closed-loop rollouts.
- Evaluation — the paradigms for diagnosing genuine memory capabilities.
The authors conclude by synthesizing open challenges, including composable and resource-aware memory architectures, trustworthy state updating, self-rollout learning, and standardized evaluation. They describe the work as the first survey on memory mechanisms for AR video generation, aiming to bridge representations, mechanisms, and learning paradigms as a structured foundation for reliable, memory-conditioned video generation systems.