Snap has unveiled an AI text-to-image mannequin for cellular gadgets that may energy a few of Snapchat’s options within the coming months. The corporate stated on Tuesday that the mannequin can produce high-resolution photos in round 1.4 seconds on an iPhone 16 Professional Max.
The brand new diffusion mannequin runs totally on the system, which reduces computational prices in comparison with fashions that depend on giant servers. Snap notes that the mannequin can create “beautiful” visible outcomes by “transferring wealthy representations from large-scale diffusion fashions to this smaller and extra environment friendly mannequin.”
Snap plans to carry the expertise into manufacturing within the coming months to assist energy the app’s AI Snaps, AI Bitmoji Backgrounds, and extra.
The corporate says that by implementing this in-house expertise, will probably be capable of supply its group high-quality AI instruments at a decrease working value.
“Snap has an extended historical past of analysis excellence in mannequin optimization and effectivity,” the corporate stated in a weblog submit. “We’re impressed by the trade innovation that’s making AI instruments extra environment friendly, inexpensive, and accessible and we stay up for persevering with to contribute to the fast tempo of innovation, notably for mobile-first experiences.”
The corporate didn’t share any additional particulars in regards to the mannequin however famous that it’s the continuation of its long-term funding in AI and ML applied sciences.
Whereas Snap has tapped AI instruments from firms like OpenAI and Google previously to energy a few of its options, it has now debuted its personal in-house AI mannequin that may enable it to supply distinctive options to its customers. It is smart for the corporate to put money into AI and develop its personal mannequin, particularly since practically each tech firm, together with Meta, has been doing so over the previous few years.
Snap is scheduled to launch its This fall 2024 earnings report on Tuesday.