AutoPP: Towards Automated Product Poster Generation and Optimization

TL;DR

AutoPP presents an end-to-end automated pipeline for product poster generation and online CTR optimization, unifying background, text, and layout design through a single multimodal design module and rendering via a token-based, decomposed-attention framework. The optimizer leverages online CTR feedback with Isolated Direct Preference Optimization (IDPO) to attribute performance gains to isolated poster elements, enabling targeted refinements. The authors introduce AutoPP1M, the largest poster dataset to date, comprising one million posters and CTR feedback, and demonstrate state-of-the-art results in both offline generation quality and online CTR improvements. This work significantly advances scalable, automated poster generation and optimization, with practical impact for marketing automation and product promotion at scale.

Abstract

Product posters blend striking visuals with informative text to highlight the product and capture customer attention. However, crafting appealing posters and manually optimizing them based on online performance is laborious and resource-consuming. To address this, we introduce AutoPP, an automated pipeline for product poster generation and optimization that eliminates the need for human intervention. Specifically, the generator, relying solely on basic product information, first uses a unified design module to integrate the three key elements of a poster (background, text, and layout) into a cohesive output. Then, an element rendering module encodes these elements into condition tokens, efficiently and controllably generating the product poster. Based on the generated poster, the optimizer enhances its Click-Through Rate (CTR) by leveraging online feedback. It systematically replaces elements to gather fine-grained CTR comparisons and utilizes Isolated Direct Preference Optimization (IDPO) to attribute CTR gains to isolated elements. Our work is supported by AutoPP1M, the largest dataset specifically designed for product poster generation and optimization, which contains one million high-quality posters and feedback collected from over one million users. Experiments demonstrate that AutoPP achieves state-of-the-art results in both offline and online settings. Our code and dataset are publicly available at: https://github.com/JD-GenX/AutoPP

Figures (11)

• Figure 1: Comparison of different poster generation and optimization pipelines. AutoPP only relies on basic product information and can optimize the generated content automatically based on CTR feedback.
• Figure 2: The framework of AutoPP. The generator synthesizes posters while the optimizer refines online CTR. For clarity, we only show the decomposed attention of MM-DiT blocks in the element rendering module. The English translations of the Chinese text are provided in parentheses.
• Figure 3: Qualitative comparison with SOTA product poster generation methods. They often suffer from issues such as obstructed products, unclear text, and incorrect text, noted by colorful boxes.
• Figure 4: Posters generated by AutoPP. Our method generates diverse layouts for various product categories.
• Figure 5: Relative CTR improvement compared with SOTA.