ImAgent: A Unified Multimodal Agent Framework for Test-Time Scalable Image Generation

TL;DR

ImAgent is a training-free, unified multimodal framework that enhances test-time image generation by integrating reasoning, generation, and self-evaluation, leading to improved fidelity and semantic alignment.

Contribution

Introduces ImAgent, a novel unified multimodal agent that operates without training to improve image generation efficiency and quality during test time.

Findings

ImAgent outperforms the backbone model in image generation and editing tasks.

It surpasses strong baselines where the backbone model fails.

Demonstrates effective test-time scaling and adaptive image generation.

Abstract

Recent text-to-image (T2I) models have made remarkable progress in generating visually realistic and semantically coherent images. However, they still suffer from randomness and inconsistency with the given prompts, particularly when textual descriptions are vague or underspecified. Existing approaches, such as prompt rewriting, best-of-N sampling, and self-refinement, can mitigate these issues but usually require additional modules and operate independently, hindering test-time scaling efficiency and increasing computational overhead. In this paper, we introduce ImAgent, a training-free unified multimodal agent that integrates reasoning, generation, and self-evaluation within a single framework for efficient test-time scaling. Guided by a policy controller, multiple generation actions dynamically interact and self-organize to enhance image fidelity and semantic alignment without relying on external models. Extensive experiments on image generation and editing tasks demonstrate that ImAgent consistently improves over the backbone and even surpasses other strong baselines where the backbone model fails, highlighting the potential of unified multimodal agents for adaptive and efficient image generation under test-time scaling.