Creating the Viral Dream App Using Amazon ECS and AWS Fargate with WOMBO

WOMBO is all about bringing apps to everyone, everywhere. “We’re the fun and accessible face of AI,” says Vivek Bhakta, WOMBO’s cofounder and head of infrastructure. The company’s mission is to make AI accessible to all so that everyone from kids to grandparents can delight in the magic of AI. WOMBO offers free apps that run on nearly any device. “You don’t need the best smartphone to run our app because we do all the processing in the cloud,” says Bhakta.

WOMBO’s biggest challenge to achieving its goals was scalability. Creating and serving AI-generated content can require 12,000 graphics processing units (GPUs) per day. “Managing all those GPUs manually would be impossible without an orchestration solution,” says Bhakta.

At first, the company was running GPUs through a different cloud provider, but as it closed in on its first app’s launch date, WOMBO realized that it needed more scalability. “We found it was very hard for us to optimize our workloads for that cloud,” says Bhakta. With an app designed to work for everyone, WOMBO needed to think ahead and plan for massive scaling. To scale its apps quickly and affordably, WOMBO migrated to AWS using Amazon ECS. “We were able to quickly adopt Amazon ECS without having to worry about configuring infrastructure,” says Bhakta. “It allows us to deploy a cluster within 15 minutes.”
 

WOMBO’s first app went massively viral in 2021, hitting 25 million downloads 3 weeks after its launch. The debut app’s success motivated the WOMBO team to develop a second app called Dream, an AI art generator that helps users create beautiful images by combining unique prompts with custom art styles. The team was able to quickly build and deploy the app using fully managed AWS solutions. “Most of our time was not spent on infrastructure but on the actual coding,” says Bhakta. “Generally, we would need a lot of time to figure out scaling and infrastructure, but in this case, it took less than 1 week.” In May 2022, Dream also went viral, supporting 5.2 million monthly active users.

With a modestly sized engineering team, WOMBO handles a massive customer base. The team launched its first app with only three engineers. At peak capacity, WOMBO operates with a dedicated team of seven engineers and four additional professionals focusing on the backend architecture. WOMBO operates with this lean, efficient team by using AWS tools strategically to run applications and manage data at peak efficiency. WOMBO runs its application workloads on serverless containers by using Amazon ECS with AWS Fargate, a serverless, pay-as-you-go compute engine, letting its engineers focus on building applications without managing servers while deploying APIs quickly. “On AWS Fargate, there’s nothing for us to manage other than deciding how many nodes we want running,” says Bhakta. This support means the WOMBO team can allocate its resources and talents toward app development, innovation, and enhancing the overall user experience.

To maintain smooth operations during viral surges, WOMBO uses AWS Lambda, a compute service that lets users run code without provisioning or managing servers. WOMBO uses a database to store all the images that Dream generates—almost two billion images since 2022. Then it uses AWS Lambda to check the database, query for tasks that don’t need to live in the database any longer, and export those tasks to a storage service. “We’re not deleting them forever,” says Bhakta. “They’re just getting saved elsewhere. And we don’t have to pay for unused compute time. It’s extremely efficient.”

By taking advantage of the efficiency of AWS services, the WOMBO team has lowered its bills. “When we moved over to AWS, we could select the right instance size for our GPUs, which helped us optimize for cost and performance,” says Bhakta. “Being able to access compute and scale up that compute on demand is extremely valuable. If we tried to launch this product without AWS, it would feel next to impossible.”

WOMBO improved price performance further when AWS released Amazon Elastic Compute Cloud (Amazon EC2) G5 Instances, the latest generation of NVIDIA GPU-based instances, which can be used for a wide range of graphics-intensive and machine learning use cases. “We realized that we’d get more than three times the output using Amazon EC2 G5 Instances,” says Bhakta. “Now, we’ve shifted our whole workload to those instances, and it has changed the way that we operate because we have better margins.”

WOMBO’s trust in the reliability of Amazon ECS has led to streamlined operations. “We don’t have to spend the time researching all the other solutions because we can trust that with Amazon ECS, we’ll be able to handle whatever we put on it and it’s going to be easy,” says Bhakta. “Our infrastructure has been largely on autopilot. Using Amazon ECS definitely saves a lot of time for us.”