cs.CVMar 14, 2026

Multi-Object Advertisement Creative Generation

Authors: Jialu Gao, Mithun Das Gupta, Qun Li, Raveena Kshatriya, Andrew D. Wilson, Keng-hao Chang, Balasaravanan Thoravi Kumaravel

Organizations: CMU, USA · Microsoft AI, USA · Microsoft Research, USA

Read PDF
Ask AI
Cite

Copy citation

DOINot available for this paper

Abstract

Lifestyle images are photographs that capture environments and objects in everyday settings. In furniture product marketing, advertisers often create lifestyle images containing products to resonate with potential buyers, allowing buyers to visualize how the products fit into their daily lives. While recent advances in Generative Artificial Intelligence (GenAI) have given rise to realistic image content creation, their application in e-commerce advertising is challenging because high-quality ads must authentically representing the products in realistic scearios. Therefore, manual intervention is usually required for individual generations, making it difficult to scale to larger product catalogs. To understand the challenges faced by advertisers using GenAI to create lifestyle images at scale, we conducted evaluations on ad images generated using state-of-the-art image generation models and identified the major challenges. Based on our findings, we present CreativeAds, a multi-product ad creation system that supports scalable automated generation with customized parameter adjustment for individual generation. To ensure automated high-quality ad generation, CreativeAds innovates a pipeline that consists of three modules to address challenges in product pairing, layout generation, and background generation separately. Furthermore, CreativeAds contains an intuitive user interface to allow users to oversee generation at scale, and it also supports detailed controls on individual generation for user customized adjustments. We performed a user study on CreativeAds and extensive evaluations of the generated images, demonstrating CreativeAds's ability to create large number of high-quality images at scale for advertisers without requiring expertise in GenAI tools.

Figures & tables

Figure 1. Overview of CreativeAds, an interactive system that supports scalable high-quality multi-product ad generation. Supplemental demo video can be found here .
Figure 2. The initial workflow design for multi-product ad generations consists of three stages: product sampling, placement, and background generation.
Figure 3. Common failure cases for directly applying state-of-the-art generation models for multi-product ad creations.
Figure 4. Overview of the system workflow. CreativeAds can batch generate high-quality ads given a user specified room-type and generation style. The system consists of three core modules, each designed to address a key challenge in realistic and authentic ad creation. First, the Product Pairing Module selects product combinations that are both semantically and visually compatible, ensuring functional coherence and viewpoint consistency. Next, the Layout Generation Module determines the spatial arrangement and relative sizes of products on a blank canvas to achieve a natural and balanced composition. Finally, the Background Generation Module inpaints the surrounding environment to highlight the selected products while preserving overall visual authenticity.
Figure 5. User Interface of CreativeAds. The interface of CreativeAds is designed to support efficient control and inspection of batch ad generation. On the left panel, users can specify high-level parameters such as the desired room type and visual style. The central panel allows users to select specific product category combinations to guide the generation process. On the right panel, a gallery view presents the generated images in a compact, scrollable layout. Users can click on individual thumbnails to view full-size images, inspect generation details, and make fine-grained adjustments to individual outputs as needed.
Figure 6. Illustration of the Product Pairing Module. This module first categorizes products into room types using a vision-language model (VLM) and then pairs semantically compatible items. To ensure visual coherence, it further filters product pairs by inferring the camera tilt from product images, selecting only those captured from similar viewpoints.
Figure 7. Examples of diverse product images from the catalog. These images are captured using different cameras, lighting conditions, and viewpoints.
Figure 8. Illustration of the Layout Generation Module. This module first generates a structured layout description prompt and extracts product dimension information from metadata to ground the layout in physical realism. It then leverages a vision-language model (VLM) to predict the spatial coordinates, relative sizes, and layering order of each product within the scene.
Figure 9. Illustration of the Background Generation Module. This module first employs a vision-language model (VLM) to generate semantically meaningful scene descriptions that guide background synthesis. It then applies masked inpainting with ControlNet to produce realistic, context-aware environments that both highlight the products and preserve their visual authenticity.
Figure 10. Scenario walkthrough 1 - Monitoring batch generation and adjusting individual ones. CreativeAds enables users to oversee batch generations organized by room types and generation styles. Additionally, it provides options for users to make adjustments to individual generations based on their specific preferences.
Figure 11. Scenario walkthrough 2 - Review. CreativeAds groups the generated images by room types, styles, and product category pairs. The results are displayed in a carousel for easy browsing, allowing for an efficient final review of the generated content.
 A1A2A3A4Ours
CLIP( ↑\uparrow )29.829.930.531.030.8
Aesthetic( ↑\uparrow )0.2590.2370.2120.2300.267
Table 1. Automatic ablation study using pre-trained models on CreativeAds.
 A1A2A3A4Ours
GPT-4o4.4104.64.3554.2754.282
GPT-4o (SoM)4.4104.7004.4844.3504.385
GPT-4.5-preview3.5643.9003.7103.1503.641
GPT-4.5-preview (SoM)3.7443.8503.9033.1753.923
Table 2. Ablation study on product authenticity.
 A1A2A3A4Ours
GPT-4o4.2564.1254.1944.3754.282
GPT-4o (SoM)4.0774.0754.0324.0754.0256
GPT-4.5-preview3.2563.3753.2903.4253.385
GPT-4.5-preview (SoM)3.0003.1502.9683.3503.103
Table 3. Ablation study on visual appeal.
 A1A2A3A4Ours
GPT-4o4.03.7753.9034.053.846
GPT-4o (SoM)3.9493.8504.0324.0253.949
GPT-4.5-preview2.4622.4752.4192.4252.385
GPT-4.5-preview (SoM)2.4362.6752.4522.6752.410
Table 4. Ablation study on layout quality.
 A1A2A3A4Ours
GPT-4o4.4624.3754.4194.6504.462
GPT-4o (SoM)4.2054.2754.2584.2754.179
GPT-4.5-preview3.58973.8253.6453.9754.103
GPT-4.5-preview (SoM)3.4363.8253.3873.9253.821
Table 5. Ablation study on theme alignment.
Figure 12. Qualitative results of ablations on sampled images. Removing the product pairing module (A1) leads to functionally mismatched products, removing the size estimation module (A2) results in unrealistic product sizes, removing the layout planning module (A3) causes unsatisfactory side-by-side alignment, and removing structural constraints during background generation (A4) results in inappropriate product appendages, violating authenticity.
Appendix figures & tables2 assets

Supplementary material from the paper’s appendix.

Appendix

Figure 13. Examples of other generated advertisements.
Figure 14. Examples of other generated advertisements.
More in this topic

Image Generation

Image generation studies models that create or edit images from text, examples, layouts, and other structured controls.

See all 463 papers

Explore similar work

CardsList
Sep 11, 2026

I Am AdMan: A Pipeline for Automatic Generation of Personalized Advertising Imagery

Personalized marketing can increase customer engagement, satisfaction, and conversion. While existing personalization approaches have become effective at matching the right product to the right customer, the visual representation of advertisements remains generic and only weakly tailored to the individual. Prior research shows that generative artificial intelligence can improve the creation of personalized advertisements, particularly for text, and that image generation models can support scalable advertisement production. However, little research has examined how detailed customer information can be systematically translated into fully AI-generated, personalized advertising imagery at scale on a technical level. To address this gap, we propose AdMan, a multi-agent pipeline that transforms customer data into personas, generates personalized advertisement images conditioned on product reference images, and applies an LLM-based judge agent for automated quality control. We implement the pipeline with two different model configurations and evaluate it across four products, using six celebrity personas for qualitative inspection, and 100 real customer profiles, producing 1745 advertisements. The evaluation combines a qualitative expert focus group and a quantitative artifact-rate assessment. The results show that the pipeline can generate photorealistic and personalized advertisements. At the same time, performance varies substantially by product complexity and model configuration. Our findings extend the literature on AI-based personalized advertising by demonstrating the feasibility and current limitations of fully automated image generation for advertising.

Victor Kolominsky-Rabas, Leopold Müller, Claudius Budcke +1

University of Bayreuth, Universitätsstraße 30, 95447 Bayreuth, Germany · Fraunhofer FIT, Wittelsbacherring 10, 95444 Bayreuth, Germany

Conditional Image GenerationImage GenerationAgentic Image GenerationPersona-Conditioned Generation
AskPDF
May 12, 2026

Design Your Ad: Personalized Advertising Image and Text Generation with Unified Autoregressive Models

Generating realistic and user-preferred advertisements is a key challenge in e-commerce. Existing approaches utilize multiple independent models driven by click-through-rate (CTR) to controllably create attractive image or text advertisements. However, their pipelines lack cross-modal perception and rely on CTR that only reflects average preferences. Therefore, we explore jointly generating personalized image-text advertisements from historical click behaviors. We first design a Unified Advertisement Generative model (Uni-AdGen) that employs a single autoregressive framework to produce both advertising images and texts. By incorporating a foreground perception module and instruction tuning, Uni-AdGen enhances the realism of the generated content. To further personalize advertisements, we equip Uni-AdGen with a coarse-to-fine preference understanding module that effectively captures user interests from noisy multimodal historical behaviors to drive personalized generation. Additionally, we construct the first large-scale Personalized Advertising image-text dataset (PAd1M) and introduce a Product Background Similarity (PBS) metric to facilitate training and evaluation. Extensive experiments show that our method outperforms baselines in general and personalized advertisement generation. Our project is available at https://github.com/JD-GenX/Uni-AdGen.

Yexing Xu, Wei Feng, Shen Zhang +15

Sun Yat-Sen University · Northeastern University

Online AdvertisingMultimodal GenerationAutoregressive Image GenerationE-Commerce
AskPDF
May 27, 2026

Utility-Aware Multimodal Contrastive Learning for Product Image Generation

Product images strongly influence consumer decision-making in online marketplaces. Empowered by multimodal contrastive learning, generative AI can output images that closely align with text prompts. Yet existing generative AI models do not directly optimize marketplace performance. This is a critical gap, since semantic alignment alone does not guarantee that an image will sell. To address this limitation, we propose a utility-aware multimodal contrastive learning framework that incorporates consumer demand into a novel Utility-Aware InfoNCE loss. Optimizing this utility-aware objective guides generation toward images that are both semantically coherent and demand-enhancing. This effect arises directly from a shift in the learned image-text representation space toward demand-driven visual cues, which we also validate through the theoretical bound of the proposed objective. In downstream applications on Amazon and Airbnb, product images generated and edited by our method outperform state-of-the-art models in increasing demand and preserving fidelity, while maintaining text-image consistency. Notably, our utility-aware framework preserves inverse U-shaped demand patterns for attributes such as aesthetics and uniqueness, improving demand-based performance while preserving fidelity and semantic consistency. Human-subject experiments further validate its commercial effectiveness. As generative AI technology continues to evolve, our utility-aware component can be flexibly embedded into emerging generative models to improve direct commercial use.

Xiaohang Feng, Yiling Xie

City University of Hong Kong

Image GenerationE-CommerceMultimodal Contrastive Learning
AskPDF