arXiv cs.AI· Aman Tyagi, Hemanth Boinpally, Jonathan Chen, Douglas Gebert, Steven Hickson·· 6 小时前AI 评分30
超越试错：图像生成视频一致性的智能体优化框架
Beyond Trial-and-Error: Agentic Optimization for Image-to-Video Adherence
AI 导读
研究者提出 Agentic Self-Improvement 框架，将黑盒 Image-to-Video（I2V）生成的试错过程重构为闭环、目标导向的优化。该框架先用多模态大语言模型（mLLM）结合 DSG 语义一致性查询与 CMQ 伪影检测迭代优化提示词，再用贝叶斯优化联合搜索随机种子与 CFG scale，并引入基于 DSG 和 CMQ 的 VTA 评分。
正文 · 原文
Abstract:Modern black-box Image-to-Video (I2V) models offer powerful capabilities in automated content creation, yet their lack of fine-grained control and reliability presents significant challenges in professional workflows. Their inherent stochasticity causes minor variations in textual prompts or hyperparameters to yield drastically different outputs often necessitating inefficient, brute-force trial-and-error processes. To address these limitations, we introduce the ``Agentic Self-Improvement" framework, which reframes video synthesis into a closed-loop, goal-directed optimization. Our framework systematically navigates the generation parameter space using a novel two-stage approach. In the first stage, an iterative prompt optimization loop uses a multimodal Large Language Model (mLLM) to refine the input prompt. This refinement implements two automated evaluations: Davidsonian Scene Graph (DSG) queries ensure semantic adherence, and Common Mistake Questions (CMQ) for artifact detection. At the second stage, we use Bayesian optimization to efficiently co-optimize stochastic seeds and CFG scales. This search is guided by a suite of quality metrics, including the novel Video-Text Adherence (VTA) score derived from the DSG and CMQ evaluations. Our framework significantly outperforms unguided search methods: in human preference studies, videos generated via our agentic approach were strongly preferred over baseline outputs, achieving win rates up to 69\%. This work provides a practical and extensible methodology for enhancing the predictability and control of state-of-the-art video generation models, moving the field beyond speculative curiosities toward reliable, production-ready tools.
| Subjects: | Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Multimedia (cs.MM) |
| Cite as: | arXiv:2608.12290 [cs.CV] |
| (or arXiv:2608.12290v2 [cs.CV] for this version) | |
| https://doi.org/10.48550/arXiv.2608.12290 arXiv-issued DOI via DataCite |
Submission history
From: Aman Tyagi [view email]
[v1]
Wed, 12 Aug 2026 17:35:16 UTC (1,635 KB)
[v2]
Mon, 26 Oct 2026 13:52:08 UTC (2,252 KB)
来源：arXiv cs.AI · arxiv.org