arXiv 2610.181422026-10-14cs.CV

FlowCred：用于联合文本到视频后训练的分层信用分配与功能协调

FlowCred: Hierarchical Credit Assignment and Functional Coordination for Joint Text-to-Video Post-Training

Shanghai Jiao Tong University（上海交通大学）；KlingAI；The Hong Kong University of Science and Technology（香港科技大学）；South China University of Technology（华南理工大学）；The University of Hong Kong（香港大学）

Mingxuan Zhu, Yanwei Huang, Shiyuan Liu, Xinzhi Zhang, Shujie Li, Yuanxing Zhang, Lu Chen

阅读 PDF

arXiv 页面 ↗HTML 原文 ↗

AI总结：

FlowCred是用于联合文本到视频后训练的两级框架，通过分层信用分配与功能协调解决联合训练的歧义，在多个T2V基准上提升性能且不增加预算，效果获多维度验证。

摘要：

提示优化和生成器后训练是改进文本到视频（T2V）生成的两种互补方式，近期系统越来越多地联合训练这两个模块。但从同一终端奖励优化提示策略和视频生成器会产生两个歧义：提示质量与随机生成相混淆，且两个模块可能实现重叠的局部改进。我们提出FlowCred，一种用于联合T2V后训练的两级框架。首先，分层信用分配使用匹配的提示-种子展开网格，在共享随机性下分离提示级信用与提示条件生成器残差。其次，功能协调在实际生成轨迹的状态下探测替代当前策略提示，为提示优化提供生成器感知证据，同时选择性衰减已被条件变化支持的生成器更新。在多个T2V基准上，FlowCred始终改进匹配的联合训练基线。在LTX-2.5上，FlowCred将严格匹配的联合训练对照在T2V-CompBench上的表现从61.47提升至64.46，在StoryEval上从55.58提升至58.08，且未增加解码视频或奖励模型调用预算。因子分析和已实现更新分析显示提示塑造与生成器路由的互补贡献，人工评估及跨骨干、更新器族的迁移提供进一步支持证据。

英文摘要：

Prompt optimization and generator post-training are two complementary ways to improve text-to-video (T2V) generation, and recent systems increasingly train both modules jointly. However, optimizing a prompt policy and a video generator from the same terminal reward creates two ambiguities: prompt quality is confounded with stochastic generation, and the two modules may realize overlapping local improvements. We introduce FlowCred, a two-level framework for joint T2V post-training. First, hierarchical credit assignment uses matched prompt-seed rollout grids to separate prompt-level credit from prompt-conditioned generator residuals under shared randomness. Second, functional coordination probes alternative current-policy prompts at states from actual generation trajectories, providing generator-aware evidence for prompt optimization while selectively attenuating generator updates already supported by conditioning changes. Across multiple T2V benchmarks, FlowCred consistently improves matched joint-training baselines. On LTX-2.5, FlowCred improves a strictly matched joint-training control from 61.47 to 64.46 on T2V-CompBench and from 55.58 to 58.08 on StoryEval, without increasing decoded-video or reward-model-call budgets. Factorial and realized-update analyses show complementary contributions from prompt shaping and generator routing, while human evaluation and transfer across backbones and updater families provide further supporting evidence.

机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。