Qualcomm demonstrates Stable Diffusion on Android phone

Qualcomm has shown a demo of AI image generator Stable Diffusion running on a Snapdragon 8 Gen 2 soc from an unknown phone. The software there generates an image of a fantasy-style cat in less than 15 seconds.

It is an image of 512 by 512 pixels, with the citeria ‘super cute fluffy cat warrior in armor, photorealistic, 4k, ultra detailed, vray rendering, unreal engine’. In the video, that led to a result after 14.42 seconds. Qualcomm says that it is the first time that Stable Diffusion works on a smartphone, but research from The Verge indicates that someone was already ahead of the company. It is true that that person took one hour in his tests with a Snapdragon 865, so in terms of efficiency Qualcomm can claim that it has a first.

In the video it shows the software in action, but in addition, the American company has shown the result three more times with the same input. In addition, it also shares the result of the prompt ‘Japanese garden at wildlife river and mountain range, highly detailed, digital illustration, artstation, concept art, matte, sharp focus, illustration, dramatic, sunset, hearthstone, art by Artgerm and Greg Rutkowski and Alphonse Mucha’. This also appears to be a product of the same hardware and software, but Qualcomm doesn’t say so emphatically. Qualcomm was able to achieve the result so quickly by making optimizations to its Qualcomm AI Stack. It also speaks of ‘tight hardware and software co-design’. His post goes into more technical details.

Stable Diffusion is an open source model that can be used to generate images based on text input. Users can run the software on a PC, unlike services like Dall-E, which are only available via the cloud. The software is supported by many GPUs with 8GB or more memory. The software is available for free on GitHub. There are also online demos of the AI ​​image generator available.