MemFlow: Flowing Adaptive Memory for Consistent and Efficient Long Video Narratives
This addresses the problem of inconsistent long video narratives for streaming video generation, representing an incremental improvement over fixed memory strategies.
The paper tackles the challenge of maintaining content consistency in long video generation by proposing MemFlow, which dynamically updates a memory bank with relevant historical frames based on text prompts, achieving strong narrative coherence with minimal computational overhead (7.9% speed reduction compared to baseline).
The core challenge for streaming video generation is maintaining the content consistency in long context, which poses high requirement for the memory design. Most existing solutions maintain the memory by compressing historical frames with predefined strategies. However, different to-generate video chunks should refer to different historical cues, which is hard to satisfy with fixed strategies. In this work, we propose MemFlow to address this problem. Specifically, before generating the coming chunk, we dynamically update the memory bank by retrieving the most relevant historical frames with the text prompt of this chunk. This design enables narrative coherence even if new event happens or scenario switches in future frames. In addition, during generation, we only activate the most relevant tokens in the memory bank for each query in the attention layers, which effectively guarantees the generation efficiency. In this way, MemFlow achieves outstanding long-context consistency with negligible computation burden (7.9% speed reduction compared with the memory-free baseline) and keeps the compatibility with any streaming video generation model with KV cache.