ImagerySearch: Adaptive Test-Time Search for Video Generation Beyond Semantic Dependency Constraints

2025-10-16

AI Summary

Video generation models perform well in realistic scenarios but degrade significantly in tasks requiring imagination—such as rare concept co-occurrence and long-distance semantic relations—due to distributional shift beyond training data. Existing test-time scaling methods employ fixed search spaces and static reward functions, lacking semantic adaptability to input prompts. To address this, we propose a prompt-guided adaptive test-time search framework that dynamically adjusts both the inference search space and reward function to enhance temporal coherence and visual plausibility. This work introduces, for the first time, an adaptive test-time search mechanism and establishes LDT-Bench—the first benchmark explicitly designed for evaluating long-distance semantic understanding in video generation. We further develop an automated evaluation protocol to quantify creative generation capability. Experiments demonstrate that our method substantially outperforms state-of-the-art baselines and existing test-time approaches on LDT-Bench, while achieving comparable gains on VBench.

Technology Category

• Natural Language Processing: Generation
• Computer Vision: Diffusion Models for Vision
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Large language models for search
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems

Abstract

Video generation models have achieved remarkable progress, particularly excelling in realistic scenarios; however, their performance degrades notably in imaginative scenarios. These prompts often involve rarely co-occurring concepts with long-distance semantic relationships, falling outside training distributions. Existing methods typically apply test-time scaling for improving video quality, but their fixed search spaces and static reward designs limit adaptability to imaginative scenarios. To fill this gap, we propose ImagerySearch, a prompt-guided adaptive test-time search strategy that dynamically adjusts both the inference search space and reward function according to semantic relationships in the prompt. This enables more coherent and visually plausible videos in challenging imaginative settings. To evaluate progress in this direction, we introduce LDT-Bench, the first dedicated benchmark for long-distance semantic prompts, consisting of 2,839 diverse concept pairs and an automated protocol for assessing creative generation capabilities. Extensive experiments show that ImagerySearch consistently outperforms strong video generation baselines and existing test-time scaling approaches on LDT-Bench, and achieves competitive improvements on VBench, demonstrating its effectiveness across diverse prompt types. We will release LDT-Bench and code to facilitate future research on imaginative video generation.

Problem

• Improving video generation for imaginative scenarios with rare concepts
• Overcoming limitations of fixed search spaces in test-time scaling
• Addressing long-distance semantic relationships beyond training distributions

Innovation

• Dynamic search space and reward adjustment
• Prompt-guided adaptive test-time search strategy
• Automated benchmark for long-distance semantic prompts