GEMS: Agent-Native Multimodal Generation with Memory and Skills

TL;DR

GEMS is a multimodal generation framework that leverages agent-based components with memory and skills to improve performance on complex tasks beyond foundational model limitations.

Contribution

It introduces a structured multi-agent framework with persistent memory and domain-specific skills to enhance multimodal generation on various tasks.

Findings

GEMS outperforms existing models on five main tasks and four downstream tasks.

A lightweight 6B model surpasses state-of-the-art larger models on GenEval2.

The framework achieves significant performance gains across multiple generative backends.

Abstract

Recent multimodal generation models have achieved remarkable progress on general-purpose generation tasks, yet continue to struggle with complex instructions and specialized downstream tasks. Inspired by the success of advanced agent frameworks such as Claude Code, we propose \textbf{GEMS} (Agent-Native Multimodal \textbf{GE}neration with \textbf{M}emory and \textbf{S}kills), a framework that pushes beyond the inherent limitations of foundational models on both general and downstream tasks. GEMS is built upon three core components. Agent Loop introduces a structured multi-agent framework that iteratively improves generation quality through closed-loop optimization. Agent Memory provides a persistent, trajectory-level memory that hierarchically stores both factual states and compressed experiential summaries, enabling a global view of the optimization process while reducing redundancy. Agent Skill offers an extensible collection of domain-specific expertise with on-demand loading, allowing the system to effectively handle diverse downstream applications. Across five mainstream tasks and four downstream tasks, evaluated on multiple generative backends, GEMS consistently achieves significant performance gains. Most notably, it enables the lightweight 6B model Z-Image-Turbo to surpass the state-of-the-art Nano Banana 2 on GenEval2, demonstrating the effectiveness of agent harness in extending model capabilities beyond their original limits.