In a groundbreaking development, Android phones may soon be equipped with the ability to generate images through artificial intelligence from simple text commands. The announcement came from Qualcomm, the leading manufacturer of cellular phone chips, who created an image of a “warrior cat” with a resolution of 512 x 512 pixels from a smartphone equipped with Google’s operating system. The photo was generated using Stable Diffusion, a popular deep learning model that converts text sentences into detailed images.
The entire process of creating the image took only 15 seconds, which is considered low for a mobile device’s hardware, even though the generated image had lower resolution. According to Qualcomm, the phone that created the warrior cat image is equipped with the Snapdragon 8 Gen 2 processor, the same processor found in top-of-the-line devices like the Samsung Galaxy S23. The chip has dedicated cores for AI activities.
To create the image, the phone received the following text prompt command: “Super cute and furry armored warrior cat, realistic photo, 4K, ultra-detailed, ray tracing, unreal engine.” According to Qualcomm, the Stable Diffusion version used in the phone was optimized to work stably on the company’s processor and was not publicly available.
Although there are already apps and websites capable of generating images with text commands, the difference is that Qualcomm’s development performed all the necessary operations for file creation from the smartphone, without an internet connection. The manufacturer did not specify which phone model was used to execute the 1.5 version of Stable Diffusion.
Creating images using a cell phone is not new. This month, developer Ivon Huang published how he used Stable Diffusion to create an image from a phone equipped with Qualcomm’s Snapdragon 865. However, the generative model required an hour to create a 512 x 512 pixel image, compared to the 15 seconds achieved in Qualcomm’s test.
The reason for developing solutions of this kind is to save costs for developers (since connection to the internet for cloud processing is unnecessary). Furthermore, it is possible to protect the user’s privacy, as the photos remain on the device. There is no information yet on when users will officially benefit from this feature, but it is expected that Qualcomm will demonstrate the technology at MWC 2023.
The introduction of AI-generated images from text commands on mobile devices can have far-reaching implications in various fields, including graphic design, art, and advertising. It could also simplify the process of image creation, making it accessible to everyone, regardless of their artistic abilities. Furthermore, the ability to create images offline using a mobile device can have significant implications for remote and low-resource areas. With this development, we may be on the verge of a new era in image creation and accessibility.