LogInThe AI can generate images in under 15 seconds.
Qualcomm AI Research has provided the "world’s first on-device demonstration of Stable Diffusion": the text-to-image tool runs on a phone and can generate pictures in under 15 seconds – a huge jump compared to the time it takes to do the same in the cloud.
The team performed full-stack AI optimizations through quantization, compilation, and hardware acceleration using the Qualcomm AI Stack to deploy Stable Diffusion on an Android smartphone powered by Snapdragon 8 Gen 2 Mobile Platform.
"Our full-stack AI research means optimizing across the application, the neural network model, the algorithms, the software, and the hardware, as well as working across disciplines within the company."
Quantization increased performance and saved power by allowing the model to consume less memory bandwidth. For compilation, Qualcomm used AI Engine direct framework to map the neural network into a program that efficiently runs on the target hardware. It improved performance and minimized memory spillage.
As a result, Stable Diffusion can create images on a smartphone in under 15 seconds for 20 inference steps and generate a 512x512 pixel picture – "this is the fastest inference on a smartphone and comparable to cloud latency".
This breakthrough means the optimizations for Stable Diffusion can be used on other platforms like laptops, XR headsets, and any other device powered by Qualcomm Technologies. The company is planning to develop the tech and broader its influence further.
Read more about it here and don't forget to join our 80 Level Talent platform, our Reddit page, and our Telegram channel, follow us on Instagram and Twitter, where we share breakdowns, the latest news, awesome artworks, and more.