Case Studies
19 March, 2024
Reface uses Google Kubernetes Engine to develop and scale its face-swapping app as it grows in popularity.
Google Cloud results
- Builds powerful machine learning models with Google Kubernetes Engine
- Scales up and down cost-effectively with Google Kubernetes Engine
- Enables Reface to evolve from an app to a social network
Scaling to make 130 million downloads with Google Cloud
We live in the age of the selfie. It’s one of the biggest cultural phenomena of the 21st century. We can show friends and family where we are, and how we’re feeling with one simple autobiographical photo. And now the selfie is evolving. Filters and editing tools let us discover a new sense of fun and creativity with our self-portraits.
One start-up asked a simple question: ‘What if you could swap your face onto anything, from a famous movie star to the Mona Lisa?’ That start-up is Reface. Its initial app was an instant success, making headlines around the world and even catching the eye of Elon Musk. Reface has since gone on to achieve over 130 million downloads and scored the number one spot on the Google Play chart.
“We chose Google Cloud because it was easy to understand. And it was evolving much faster than other cloud providers. Google Cloud could match the speed at which we were working, and it evolves faster and faster each day.” —Maksym Piddubnyi, Head of IT, Reface.
But, as with many successful start-ups, the initial phase of turning a simple, saleable idea into something more required a lot of hard work, agile thinking, and improvisation. Google Cloud has been a part of Reface since its early days, its founders say that the platform offers ease of use and a constantly evolving suite of products.
“We chose Google Cloud because it was easy to understand,” says Maksym Piddubnyi, Head of IT at Reface. “And, it was evolving much faster than other cloud providers. Google Cloud could match the speed at which we were working, and it evolves faster and faster each day. We were also attracted by the ethical business approach of Google, as well as the range of services Google Cloud has to offer, such as Google Kubernetes Engine and Pub/Sub, which was one of the first tools we started to work with.”
Challenging the Face of Selfies with Google Cloud
Changing the face of selfies with Google Cloud. Originally dubbed Doublicat, a play on the word duplicate, Reface was created by a group of Ukrainian friends who were experimenting with machine learning and artificial intelligence. Their initial efforts resulted in Reface. This allowed users to swap their faces with everyone from movie stars to celebrities to sportspeople and to create their own shareable gifs, a technology known as ‘synthetic media.’ The success of the app saw Doublicat grow from a handful of enthusiasts to become Reface, which currently has a 260-strong team. But Maksym Piddubnyi says that the spirit of adventure that led to its creation, as well as its culture of coding, has been maintained to this day.
As a young start-up, the Reface team were working fast, constantly creating and testing new iterations of their app. The company has been using on-premise infrastructure for some of its tasks, but launching the app to a global audience required scalable cloud infrastructure that could deliver results from Day 1. Having started using Google Cloud and seeing the first results, they worked together with Terasky, a Google Cloud Partner, to help with their growing cloud-based infrastructure.
“TeraSky is a great partner,” says Piddubnyi. “The whole team helped us a great deal. We have a lot in common and they’re on the same page as us, which makes communicating with them really easy. It’s always been great to work with them, and lean on their expertise to overcome any of the challenges we faced, particularly early on.”
Building face-swapping technology with Google Kubernetes Engine
The Reface app’s face-swapping effects are powered by machine learning frameworks known as generative adversarial networks (GANs). These generate a new animated face using the selfie and the target video rather than trying to mask one on top of the other. Google Kubernetes Engine gives Reface the power to build and run the high-performance infrastructure that makes this possible.
“Google Kubernetes Engine is the main place for our production infrastructure and has helped us cut big, often complex, ‘monster’ deployments into microservices. Without it, we’d struggle to deploy Docker containers or VM instance groups, for example.” —Maksym Piddubnyi, Head of IT, Reface.
The popularity of the Reface app meant that Reface needed a cloud solution that could scale quickly. Its on-premise architecture could not deal with the demands of millions of people using the app to transform their selfies, but Google Kubernetes Engine can take the strain and allows the team to deal with spikes in traffic.
“Google Kubernetes Engine is the main place for our production infrastructure, and has helped us cut big, often complex, ‘monster’ deployments into microservices,” Piddubnyi explains. “Without it, we’d struggle to deploy Docker containers or VM instance groups, for example. We knew that we wanted to work with Kubernetes clusters from the start, but we really liked the approach taken with Google Kubernetes Engine.”
To deal with its database demands, Reface uses Cloud Storage as a scalable, easy-to-use, and highly secure location to hold static assets of all kinds. It also employs Cloud SQL as a fully managed relational database service. These help Reface automate database provisioning, storage capacity management, and other time-consuming tasks. “Cloud SQL deals effectively with all our internal information, such as the results from our machine-learning models, which is a massive amount of data,” Piddubnyi adds. “With so much data being used and stored, we also use Cloud VPN as part of our security procedures.”
Experimenting with new ideas using Vertex AI
Reface has big plans for developing the Reface app and creating more features that will take it beyond a gif, video, and image creator, and into a thriving social media community. And, the Reface Research and Development (R&D) team has been using Vertex AI to explore and experiment with new machine-learning models.
Building the face of the future
Reface is looking toward a future that’s about more than just swapping faces. It wants to use Google Cloud to explore the true potential of ‘synthetic media’ technology. “We want to take something magic and package it into something simple that the end-user can understand, with the power of Google Cloud we know that we can build a selfie community with added face value,” concludes Piddubnyi. “We are constantly looking into Google Cloud developments to understand if there is something new that could be interesting for us.”
“We want to take something magic and package it into something simple that the end-user can understand, with the power of Google Cloud we know that we can build a selfie community with added face value.” —Maksym Piddubnyi, Head of IT, Reface