VideoCoF：基于时序推理器的统一视频编辑
VideoCoF: Unified Video Editing with Temporal Reasoner
把「先预测编辑区域隐变量、再生成目标视频」的思维链式两阶段生成范式引入视频扩散模型，用推理 token 替代用户掩码实现指令到区域的精确对齐；其 RoPE 对齐策略还可用于超出训练时长的视频长度外推，对做指令驱动编辑与长视频生成的人可直接复用。
原文依据 compelling the video diffusion model to first predict reasoning tokens (edit-region latents) before generating the target video tokens
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论现有视频编辑方法面临一个关键权衡：专家模型精度高，但依赖掩码等任务特定先验，难以统一；相反，统一的时序上下文学习模型无需掩码，却缺乏显式空间线索，导致指令到区域的映射较弱、定位不精确。为解决这一冲突，我们提出 VideoCoF，一种受思维链推理启发的新型「帧链」方法。VideoCoF 通过迫使视频扩散模型在生成目标视频 token 之前先预测推理 token（编辑区域隐变量），从而强制执行「先看、再推理、后编辑」的流程。这一显式推理步骤无需用户提供掩码，同时实现了精确的指令到区域对齐与细粒度视频编辑。此外，我们引入一种 RoPE 对齐策略，利用这些推理 token 来保证运动对齐，并实现超出训练时长的长度外推。我们证明，仅以 5 万对视频的最小数据成本，VideoCoF 就在 VideoCoF-Bench 上取得了最优性能，验证了该方法的效率与有效性。我们的代码、权重与数据见 https://github.com/knightyxp/VideoCoF。
英文原文摘要 1267 字符
Existing video editing methods face a critical trade-off: expert models offer precision but rely on task-specific priors like masks, hindering unification; conversely, unified temporal in-context learning models are mask-free but lack explicit spatial cues, leading to weak instruction-to-region mapping and imprecise localization. To resolve this conflict, we propose VideoCoF, a novel Chain-of-Frames approach inspired by Chain-of-Thought reasoning. VideoCoF enforces a "seeing, reasoning, then editing" procedure by compelling the video diffusion model to first predict reasoning tokens (edit-region latents) before generating the target video tokens. This explicit reasoning step removes the need for user-provided masks while achieving precise instruction-to-region alignment and fine-grained video editing. Furthermore, we introduce a RoPE alignment strategy that leverages these reasoning tokens to ensure motion alignment and enable length extrapolation beyond the training duration. We demonstrate that with a minimal data cost of only 50k video pairs, VideoCoF achieves state-of-the-art performance on VideoCoF-Bench, validating the efficiency and effectiveness of our approach. Our code, weight, data are available at https://github.com/knightyxp/VideoCoF.