Prisma Labs is a mobile technology company that specializes in deep learning and related products. Its Prisma Inference Engine allows convolutional neural networks to run on smartphones, fully utilizing the available GPU power and performing much faster than TensorFlow or Caffe. Prisma is a popular photo editing application from Prisma Labs that makes use of neural networks and artificial intelligence to enhance and beautify images.
Prisma Labs aims to bring an absolutely new and unique way for people to express their emotions and transfer them into digital creations. The team is working on photo and video style transfer as well as object detection and segmentation on devices. Their goal is to place the most advanced technology in the hands of users worldwide.
Prisma Labs works extensively with companies that produce mobile devices, cameras as well as processors to make its technology faster while consuming less power.
Holding the Reins
Artem (Tom) Semyanov, CEO of Prisma Labs, is its former COO and Director of R&D. Tom joined the Prisma team at the beginning of 2017. He began building the R&D team and infrastructure to lay the groundwork for new technologies and features. Last year, the R&D team grew to 35 engineers and researchers.
Alexey Moiseenkov, founder and the former CEO of Prisma transferred control of the company to Tom after Prisma switched to a B2B strategy. The Prisma app was developed by Alexey Moiseenkov, Ilya Frolov, Oleg Poyaganov, and Andrey Usoltsev. It was launched in June 2016 as a free mobile app.
Prisma Labs develops a wide range of technologies based on AI and Computer Vision. Its image effects allow companies to add value to their cameras while contributing to an improved user experience.
The team also does a lot of research into AR (augmented reality) technology. “We create new technology based on AR because we believe that it can transform the user experience in a way that no other solutions can,” says Tom.
The company’s own flagship technology, Style Transfer is used in the Prisma application. Tom explains, “We have our own Style Transfer architecture that gives us an opportunity to make it faster and even perform in real-time. We already offer more than 100 different exciting styles.”
Portrait segmentation is another cutting-edge technology that the company utilizes. The Prisma Labs team has developed a neural network that can separate a person from the picture background. Portrait segmentation can be utilized in Bokeh mode as well. The effect uses portrait segmentation technology for accuracy, and then blurs the background to make the picture more effective. In new premium devices such as the iPhone X and Pixel 2, Portrait mode is very popular among users; it blurs the background and creates a professional camera effect with low aperture settings. With Portrait Segmentation, users can recreate the same effect on any device and in real-time.
Earlier this year, Prisma Labs released revolutionary new technology that is essential for the AR experience. Their AR Synchronization System allows multiplayer in AR. Users can play multiplayer AR games (shoot a fireball at a friend!), create collaborative AR art such as 3D digital drawings in a shared environment, create custom markers in the environment and share them with other users (for example, digital pop-up instructions to show hotel guests how to use a safe or shower), and locate city sights using additional GPS synchronization, among other amazing abilities.
Create the Best
Prisma Labs creates the best technologies every day that perform with high accuracy, high processing speed, and low power consumption. Prisma Labs has a very skilled and motivated team spread across three international offices in Moscow, Shanghai, and Sunnyvale.
Prisma is the first company that switched from cloud-based solutions to on-device computing for neural networks. “We continue to create new optimizations for our algorithms and work closely with processor producers to provide the most efficient solutions,” asserts Tom. Prisma is among the first companies that started to use GPU acceleration device to process images. In fact, it started the trend of on-device AI processing.
Prisma Labs always knew that they had a quality product but were still surprised at how well it was received. Initially, the team had focused on creating a fast algorithm when creating the app. They then put the app on the App Store to run some tests but it suddenly went viral.
People started telling each other about an amazing new app which could turn any dull picture into a work of art within seconds. Prisma became the most downloaded app in 10 countries within nine days.
The processing at first was done on servers. “We were not ready for such rapid instant growth; we had to double the server capacity daily but even that was not enough. People were waiting 10-20 times longer than we had planned to get a picture processed. Having enough capacity was not cheap either, and that’s why we decided to do our what was best for our users and make Prisma offline,” reveals the team.
Tom adds, “And we have lots of exciting features coming! Stay tuned!”