MemRoPE: Training-Free Infinite Video Generation via Evolving Memory Tokens

TL;DR

MemRoPE introduces a training-free method for infinite video generation that maintains long-term and short-term memory streams to improve temporal coherence and fidelity over extended durations.

Contribution

It proposes a novel, training-free framework with dynamic memory tokens and positional indexing to enhance long-horizon video generation without fidelity loss.

Findings

Outperforms existing methods in temporal coherence

Improves visual fidelity in long-duration videos

Maintains subject consistency over extended periods

Abstract

Autoregressive diffusion enables real-time frame streaming, yet existing sliding-window caches discard past context, causing fidelity degradation, identity drift, and motion stagnation over long horizons. Current approaches preserve a fixed set of early tokens as attention sinks, but this static anchor cannot reflect the evolving content of a growing video. We introduce MemRoPE, a training-free framework with two co-designed components. Memory Tokens continuously compress all past keys into dual long-term and short-term streams via exponential moving averages, maintaining both global identity and recent dynamics within a fixed-size cache. Online RoPE Indexing caches unrotated keys and applies positional embeddings dynamically at attention time, ensuring the aggregation is free of conflicting positional phases. These two mechanisms are mutually enabling: positional decoupling makes temporal aggregation well-defined, while aggregation makes fixed-size caching viable for unbounded generation. Extensive experiments validate that MemRoPE outperforms existing methods in temporal coherence, visual fidelity, and subject consistency across minute- to hour-scale generation.