AI-Powered Ad Design: Alibabas PAID Framework Revolutionizes E-commerce Advertising

Friday 14 March 2025


The quest for AI-generated ads that actually work has been a long and arduous one. For years, researchers have been trying to crack the code on creating visual content that’s both compelling and effective. A new paper out of Alibaba Group takes a significant step forward in this endeavor, introducing a framework called Product-Centric Advertising Image Design (PAID).


At its core, PAID is an AI system designed to generate high-quality advertising images for e-commerce products, given just a few pieces of information: the product’s foreground image, marketing taglines, and target size. The system consists of four stages: prompt generation, layout generation, background image generation, and graphics rendering.


The first stage, prompt generation, uses a visual language model to create adaptive prompts based on the product foregrounds. This ensures that the resulting ads are tailored to the specific product being promoted. Next, the layout generation module predicts how graphic and non-graphic elements should be placed according to the prompt, product foreground, and taglines. This is where things get really clever – PAID can even generate backgrounds that match the product’s style.


The background image generation stage uses a layout-controlled inpainting model to fill in any gaps or holes in the image. Finally, the graphics rendering module combines all the generated elements into a cohesive final ad.


To train and test PAID, researchers created two large datasets: PITA (Product Image Tagline Annotation) and PIL (Product Image Layout). The PITA dataset includes advertising images with marketing taglines, while PIL contains product images without taglines. By analyzing these datasets, the team was able to fine-tune their model’s performance.


The results are impressive – PAID-generated ads outperform baseline models in terms of visual quality and spatial control. In online A/B tests, PAID increased click-through rates by 3% compared to traditional methods. This is a significant finding, as effective ad design can make all the difference in driving sales and conversions.


But what really sets PAID apart is its ability to generate high-quality ads without requiring human input or creative direction. This makes it an attractive solution for e-commerce companies looking to streamline their advertising efforts while still producing compelling visuals.


Of course, there are still limitations to PAID – for example, the system may struggle with products that have complex shapes or unusual textures. However, this is a promising first step towards creating AI-generated ads that can truly compete with human-designed content.


Cite this article: “AI-Powered Ad Design: Alibabas PAID Framework Revolutionizes E-commerce Advertising”, The Science Archive, 2025.


Ai-Generated Ads, E-Commerce Products, Advertising Images, Alibaba Group, Paid, Product-Centric, Visual Content, Marketing Taglines, Target Size, Graphics Rendering, Click-Through Rates.


Reference: Hongyu Chen, Min Zhou, Jing Jiang, Jiale Chen, Yang Lu, Bo Xiao, Tiezheng Ge, Bo Zheng, “PAID: A Framework of Product-Centric Advertising Image Design” (2025).


Leave a Reply