Qualcomm has just demonstrated a significant advance in Stable Diffusion technology, an artificial intelligence model trained on a large scale of data capable of generating images from text. For the first time, the company showed this feature running natively on a cell phone, eliminating any cloud processing.
Stable Diffusion was first demonstrated on an Android phone powered by Snapdragon 8 Gen 2. The video below shows an actual capture of a device generating the image of a “cat wearing armor” in less than 15 seconds. Check out:
To prove the capabilities of artificial intelligence, Qualcomm used the same input text to generate four images with a resolution of 512 x 512 pixels. The parameters entered for the demonstrations were the following: “super cute warrior cat in armor, photorealistic, 4K, ultra-detailed, vray rendering, unreal engine”.
Most artificial intelligence applications for image generation have cloud-based processing, that is, they use servers equipped with incredibly powerful processors to deliver a result in a matter of minutes. In Qualcomm’s demo, all processing took place on the phone itself.
doctor Jilei Hou, vice president of engineering at Qualcomm, says it took several optimizations to make Stable Diffusion work smoothly on a phone with hardware already available in the consumer market. For this, adaptations were made based on the open source version 1-5 of Stable Diffusion.
Qualcomm highlights the close ties between hardware and software, allowing the model to take full advantage of the Hexagon Processor, a processor dedicated to artificial intelligence tasks present in the AI Engine of the Snapdragon 8 Gen 2.
tech
23 Feb
tech
15 Feb
“The result of this total optimization is running Stable Diffusion on a cell phone in less than 15 seconds with just 20 steps of inference to generate an image of 512 x 512 pixels”, comments Hou, highlighting that this is the speed of processing of artificial intelligence fastest on a smartphone.
The beginning of 2023 seems to be marked by advances in artificial intelligence of different types. In addition to the image generation technology demonstrated by Qualcomm, OpenAI kicked off the popularization of machines capable of generating text with ChatGPT, which powers Microsoft’s new (and controversial) Bing Chat.
