---
title: "Beyond Trial-and-Error: Agentic Optimization for Image-to-Video Adherence"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2608.12290v1"
permalink: "https://feed.astralor.com/items/itm_2934e879cfa1627597e84127"
published: "2026-08-12T17:35:16.000Z"
score: 76
channel: "ai"
tags: ["Agentic Self-Improvement","arXiv","Bayesian Optimization","Image-to-Video","Prompt Optimization"]
---

# Beyond Trial-and-Error: Agentic Optimization for Image-to-Video Adherence

## AI commentary

> This paper tackles the critical pain point of uncontrollability in video generation, closing the loop between prompt and parameter optimization. Especially useful for teams building video generation products or seeking to reduce manual tuning.

## At a glance

Modern black-box Image-to-Video (I2V) models offer powerful automated content creation, yet lack fine-grained control and reliability due to stochastic outputs. This paper introduces an 'Agentic Self-Improvement' framework that reframes video synthesis as closed-loop, goal-directed optimization. Stage one iteratively refines prompts using a multimodal LLM with Davidsonian Scene Graph (DSG) queries for semantic adherence and Common Mistake Questions (CMQ) for artifact detection. Stage two applies Bayesian optimization to co-tune stochastic seeds and CFG scales, guided by quality metrics including a novel Video-Text Adherence (VTA) score. In human preference studies, agentic videos achieved win rates up to 69%, significantly outperforming unguided search. This work offers a practical method for enhancing predictability and control, moving video generation toward reliable production-ready tools.

## Key points

- Introduces an Agentic Self-Improvement framework that frames I2V generation as closed-loop optimization, jointly refining prompts and generation parameters in two stages.
- Novel use of DSG queries and CMQ detection with a new VTA score enables automated evaluation of semantic adherence and artifacts.
- Human preference tests show win rates up to 69%, significantly outperforming unguided search methods.

## Full text

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.

---

Verify against the original before citing; curated by Astral Feed: https://feed.astralor.com/items/itm_2934e879cfa1627597e84127