PosterAgent: Agentic Poster Generation via Stage-Aware Reinforcement Learning

Poster generation is a complex task demanding a harmonious integration of visual aesthetics and information hierarchy. While recent text-to-image models have advanced visual synthesis, they remain non-editable and struggle with precise text rendering. Conversely, existing layout-generation methods offer structure but typically rely on static, one-shot predictions, lacking the mechanism for self-correction essential to professional design. Inspired by the iterative workflow of human designers, we introduce PosterAgent, a novel framework that reformulates poster creation as an agentic workflow involving initial drafting followed by iterative refinement. To effectively train this multi-turn capability, we propose Stage-Aware Reinforcement Learning (SARL), which decouples the optimization into draft-specific and refinement-specific phases, ensuring precise credit assignment for both initial drafting and incremental refinement actions. Extensive experiments demonstrate that PosterAgent significantly outperforms strong baselines, validating the potential of agentic systems in graphic design.

Designing a good poster is hard: you must balance eye-catching visuals with a clear, organized flow of information. Recent AI image generators can paint attractive pictures, but they often garble the text and produce images you can't edit afterward. Other methods build an editable blueprint first, yet they commit to a single design in one shot, leaving no way to step back and fix their own mistakes, unlike a human designer who sketches, looks, and revises. We built PosterAgent, an AI that works the way a designer does: it drafts a layout, renders it into a real poster, critiques its own result, and makes targeted improvements over several rounds. Teaching this was tricky, because judging only the final poster blurs two different skills: making a strong first draft versus improving one. Our training method, Stage-Aware Reinforcement Learning, scores these two skills separately, so the AI learns each one properly. Despite being far smaller than systems like GPT-5, PosterAgent matches them and renders text especially well. More broadly, it shows that AI which critiques and revises its own work can tackle creative design, an approach that could extend to slides, web pages, and other layouts.