amazon.com/lp/build-your-business-with-amazon-advertising?tag=googhydr-20&ref=pd_sl_37ga81juc3_e_ps_gg_b_us_en_d_core_e_644030686063&k_amazon%20ads&group_144796814053″ target=”_blank” rel=”noopener”>amazon Ads helps advertisers and brands achieve their business goals by developing innovative solutions that reach millions of amazon customers at every stage of their journey. At amazon Ads, we believe that what makes advertising effective is delivering relevant ads in the right context and at the right time within the consumer's purchasing process. To that end, amazon Ads has used artificial intelligence (ai), applied science, and analytics to help its clients drive desired business outcomes for nearly two decades.
In a March 2023 survey, amazon Ads found that among advertisers who were unable to create successful campaigns, nearly 75 percent cited creative content creation as one of their biggest challenges. To help advertisers address this challenge more seamlessly, amazon Ads launched an imaging capability that quickly and easily develops lifestyle images, helping advertisers bring their brands' stories to life. amazon/amazon-ads-ai-powered-image-generator” target=”_blank” rel=”noopener”>This blog post shares more information about how amazon Ads generative ai solutions help brands create more visually rich consumer experiences.
In this blog post, we describe the architectural and operational details of how amazon Ads implemented its ai-powered generative imaging solution on AWS. Before delving into the solution, we start by highlighting an advertiser's creative experience enabled by generative ai. Below we present the solution architecture and process flows for creating, deploying, and inferring machine learning (ML) models. We finish with the lessons learned.
Advertiser creative experience
When creating ad creatives, advertisers prefer to personalize the creative so that it is relevant to their intended audiences. For example, an advertiser could have static images of their product on a white background. From the advertiser's point of view, the process is managed in three steps:
- Image generation converts unique product images into rich, contextually relevant images using generative ai. The approach preserves the original features of the product and does not require technical expertise.
- Anyone with access to the amazon Ads console can create custom brand images without requiring technical or design knowledge.
- Advertisers can create multiple engaging and contextually relevant product images at no additional cost.
A benefit of the image generation solution is the automatic creation of relevant product images based solely on product selection, without the need for additional input from advertisers. While there are options to enhance background images, such as prompts, themes, and custom product images, they are not necessary for engaging creative. If advertisers do not provide this information, the model will infer it based on the information in their product listing in amazon.com” target=”_blank” rel=”noopener”>amazon.com.
Figure 1. An example of the imaging solution showing a hydraulic flask with multiple bottoms.
Solution Overview
Figure 2 shows a simplified solution architecture for model inference and implementation. The steps for model development and deployment are shown in blue circles and represented with Roman numerals (i,ii,… iv.), while the inference steps are in orange with Hindu-Arabic numerals (1,2,… 8.).
Figure 2. Solution architecture for model inference and implementation.
amazon SageMaker is at the center of model development and deployment. The team used amazon SageMaker JumpStart to quickly create prototypes and iterations under the desired conditions (step i). Acting as a model hub, JumpStart provided a large selection of base models and the team quickly ran their benchmarks on the candidate models. After selecting candidate large language models (LLMs), science teams can continue with the remaining steps by adding more customization. amazon Ads Applied Scientists use SageMaker Studio as a web-based interface to work with SageMaker (step ii). SageMaker has the appropriate access policies to view some results from the intermediary model, which can be used for further experimentation (step iii).
The amazon Ads team manually reviewed images at scale through a human-in-the-loop process where the team ensured the app provided high-quality, responsible images. To do that, the team implemented test endpoints using SageMaker and generated a large number of images covering various scenarios and conditions (step iv). In this case, amazon SageMaker Ground Truth allowed machine learning engineers to easily create the human workflow in the loop (step v). The workflow allowed the amazon Ads team to experiment with different base models and configurations using blind A/B testing to ensure that feedback on the generated images is unbiased. Once the chosen model is ready to go to production, the model is deployed (step vi) using the team's internal Model Lifecycle Manager tool. Essentially, this tool uses artifacts generated by SageMaker (step vii) which is then deployed to the AWS production account (step viii), using SageMaker SDK .
Regarding inference, customers using amazon Ads now have a new API to receive these generated images. amazon API Gateway receives the PUT request (step 1). The request is then processed by AWS Lambda, which uses AWS Step Functions to orchestrate the process (step 2). The product image is pulled from an image repository, which is part of an existing solution prior to this creative feature. The next step is to process customers' text prompts and personalize the image through content ingestion guardrails. amazon Comprehend is used to detect unwanted context in the text message, while amazon Rekognition processes images for content moderation purposes (step 3). If the entries pass inspection, then the text continues as a message, while the image is processed by removing the background (step 4). Then, the implemented text-to-image model is used for image generation using the message and the processed image (step 5). The image is then uploaded to an amazon Simple Storage Services (amazon S3) image bucket, and metadata about the image is stored in an amazon DynamoDB table (step 6). This entire process starting from step 2 is orchestrated by AWS Step Functions. Finally, the Lambda function receives the image and metadata (step 7) which is then sent to the amazon Ads customer service via API Gateway (step 8).
Conclusion
This post introduced the technical solution for amazon Ads' ai-powered imaging solution, which advertisers can use to create custom brand images without the need for a dedicated design team. Advertisers have a number of features to generate and customize images, such as writing text messages, selecting different themes, swapping the featured product, or uploading a new product image from their device or asset library, allowing them to create impactful images for advertise your products. .
The architecture uses modular microservices with separate components for model development, registration, model lifecycle management (which is an orchestration-based solution and tiered functions to process advertiser input), selecting the appropriate model, and Track work throughout the service and a client. versus API. Here, amazon SageMaker is at the center of the solution, from JumpStart to the final SageMaker deployment.
If you plan to build your generative ai application in amazon SageMaker, the fastest way is with SageMaker JumpStart. See this amazon-SageMaker-JumpStart_2022_VW_s44e03-MCL_OD.html” target=”_blank” rel=”noopener”>presentation to learn how you can start your project with JumpStart.
About the authors
Anita Laca is the single-threaded leader in ai generative image ads on amazon, allowing advertisers to create visually stunning ads with the click of a button. Anita combines her extensive experience in the hardware and software industry with the latest innovations in generative ai to develop efficient, cost-optimized solutions for her clients, revolutionizing the way businesses connect with their audiences. She is passionate about traditional visual arts and is an exhibiting printmaker.
Burak Gozluklu is a Principal Solutions Architect specializing in ai/ML located in Boston, MA. He helps strategic clients adopt AWS technologies and specifically generative ai solutions to achieve their business objectives. Burak holds a PhD in Aerospace Engineering from METU, a Master's degree in Systems Engineering, and a postdoc in System Dynamics from MIT in Cambridge, MA. Burak remains a research affiliate at MIT. Burak is passionate about yoga and meditation.
Christopher of beer is a Senior Software Development Engineer at amazon located in Edinburgh, United Kingdom. He with experience in visual design. He works in the creative construction of products for advertising, focusing on video generation, helping advertisers reach their customers through visual communication. Create products that automate creative production, using traditional and generative techniques, to reduce friction and delight customers. In addition to his work as an engineer, Christopher is passionate about human-computer interaction (HCI) and interface design.
Yashal Shakti Kanungo He is an applied scientist III at amazon Ads. His focus is on fundamental generative models that take a variety of user inputs and generate text, images and videos. It is a combination of research and applied science, constantly pushing the boundaries of what is possible in generative ai. Over the years, he has researched and implemented a variety of these models in production across the spectrum of online advertising, from ad sourcing, click prediction, headline generation, image generation, and further.
Srivan Sripada is a Senior Applied Scientist at amazon located in Seattle, WA. His main focus lies on developing generative ai models that allow advertisers to create engaging advertising creatives (images, videos, etc.) with minimal effort. Previously, he worked on using machine learning to prevent fraud and abuse on amazon's retail platform. When he is not at work, he is passionate about outdoor activities and spending time meditating.
Cathy Willcock is a Senior Technical Business Development Manager located in Seattle, WA. Cathy leads the AWS technical accounts team supporting the adoption of AWS cloud technologies in amazon Ads. Her team works on amazon Ads enabling the discovery, testing, design, analysis and deployment of AWS services at scale, with a particular focus on innovation to shape the landscape in the AdTech and MarTech industry. Cathy has led engineering, product and marketing teams and is the inventor of ground-to-air calls (1-800-RINGSKY).
