FrameMorrow：面向长时程视频生成的前瞻性帧选择

AI 摘要

该论文提出 FrameMorrow，一种面向长时程视频生成的前瞻性帧选择器。现有方法通常依据当前内容判断历史帧的相关性，但作者认为历史信息应按未来信息需求来选择。FrameMorrow 通过预测一小组代表未来信息需求的前瞻 token 来筛选历史帧，且选择的是显式历史帧而非模型内部状态，因此可即插即用地集成到包括闭源模型在内的多种生成器中，额外推理成本很低。论文在五个基准和 11 个生成模型上评估，覆盖长视频生成、交互式生成和动作条件世界模型，结果显示长程一致性、视觉质量和动作对齐均有稳定提升。

正文节选

FrameMorrow: Future-guided Frame Selection with Prospective Tokens for Long-Horizon Video Generation Abstract Long-horizon video generation requires models to effectively leverage an increasingly long generation history. As the generated history grows, retaining all previous content becomes increasingly expensive and redundant, making effective historical selection essential. Existing approaches often determine historical relevance based on the current content. However, information relevant to t

发布时间：—

抓取时间：2026-10-05 13:02

来源机构：Hugging Face