"It's a really cool thing."īut why does a photo app need huge processing power like this? "It's a lot more scalable if it's on the device," said Gatys. The 27-year-old PhD student from Germany is studying machine learning and computational neuroscience.īy making Prisma more efficient so a smartphone can run the app, the team can now let many more people use it. "It's a non-trivial engineering challenge," Gatys told CNNMoney. This was accomplished by reducing the deep learning neural network "to throw away unnecessary parts" while still maintaining the performance on a weaker machine. To solve this problem, the Prisma team has, in essence, outsourced that computational process to the smartphone, according to Leon Gatys, whose research and DeepArt.io project inspired the app. With over 35,000 photos converted each minute, the Prisma team needs thousands of graphics processors, which isn't scalable. Related: New AI can predict when two people will kissĭeep learning requires a computer with 60 times the graphics processing power of a smartphone to edit one photo, according to Prisma. For a computer to understand this information all at once, many calculations are needed to figure those details out. When we see a book, we know instantaneously what the title is, its color and length. It's designed, typically, to make computers evaluate visual information like humans do. "Now, people can carry that power in their pockets."ĭeep learning requires an interconnected bundle of math formulas that compute tons of problems simultaneously.
"The technology behind Prisma - deep learning - is a bridge between your imagination and your digital creation," CEO Alexey Moiseenkov said in a statement. Prisma transforms photos into various styles of artwork.