Android phone "runs Stable Diffusion" and sets a new record, producing images within 15 seconds

Android phone "runs Stable Diffusion" and sets a new record, producing images within 15 seconds

As we know, Stable Diffusion is a very popular text-to-image generative AI model that can create realistic images for any given input text in tens of seconds. Stable Diffusion has more than 1 billion parameters, and until now, the model has mainly run in the cloud.

Therefore, how to run Stable Diffusion on mobile devices has attracted great interest from people in the industry. Previously, an author developed an application that generates images through Stable Diffusion and then edits them the way you like. The application only takes one minute to generate images on the latest iPhone 14 Pro and uses about 2GiB of application memory.

Even Apple has optimized it, running Stable Diffusion at an amazing speed on devices such as iPhone, iPad, and Mac, generating a 512x512 resolution image in half a minute. The following is an image generated by running Stable Diffusion and Core ML + diffusers on an Apple device.

Stable Diffusion achieves the fastest Android operation

So can Android phones also run Stable Diffusion and quickly generate images? Today, a demo video from Qualcomm demonstrated that Android phones can use Stable Diffusion to achieve the fastest local AI image generation.

In the demo video below, Stable Diffusion v1.5 generates a 512x512 resolution image in less than 15 seconds . Although Qualcomm did not specify which phone it was, it said that the phone was equipped with the latest Snapdragon 8 Gen2 flagship chipset. Various custom optimizations were made at the software level to enable Stable Diffusion to run in the best possible way.

Some example images Qualcomm generated on its test devices include the following: “Super cute, furry warrior cat in armor.”

Running Stable Diffusion at such a fast speed on smartphones is thanks to Qualcomm's full-stack AI optimization solution, which significantly reduces runtime latency and energy consumption, ultimately enabling the safe and efficient operation of Stable Diffusion.

In fact, before Qualcomm released the demo video, a developer had already demonstrated running Stable Diffusion on the 8G RAM Sony Xperia 5 II equipped with Qualcomm Snapdragon 865. The developer's name is Ivon Huang, and his blog details the relevant tutorials. However, it takes 1 hour to generate an image with a resolution of 512x512. In comparison, Qualcomm has achieved a huge improvement in the speed of outputting images.

Blog address:

​https://ivonblog.com/en-us/posts/android-stable-diffusion/​

<<:  Android 13 QPR2 Beta 3.2 released: fix screen artifact issues

>>:  Do you know Web3?

Recommend

Business logic splitting based on micro frontend

1. What is a micro frontend? Micro frontends are ...

How to design an App login module

I have previously explained and discussed with yo...

Explore the mystery of lunar soil during the Mid-Autumn Festival

Chairman Mao wrote in the poem "Shui Tiao Ge...

Inventory of useless features of smart TVs

Looking back on the past, the days when televisio...

The standard answers to programmer interviews are not standard

[[165225]] Peter Verhas asked a seemingly irrelev...

12 tips | Key points you need to know about creating advertising slogans

I won’t accept any gifts this year, and if I do, ...

Apple releases iOS 11.4 beta 2 with more than just ClassKit

Recently, Apple released iOS 11.4 beta 2. Two wee...

Are Brands Misreading Metaverse Marketing?

In April last year, the launch of BAYC Bored Ape ...

From a scientific point of view, how old can a person live at most?

December 16, 2021, is an unusual day because Alim...