Local image generation on smartphones seems to be soon possible thanks to a new technology developed by Qualcomm. This company, specializing in SoCs for smartphones and laptops, has created ControlNet, an artificial intelligence program that operates directly on an Android smartphone. By using a source image and a textual description as input, ControlNet is capable of generating a new image in just 12 seconds.
The demonstration of this technology took place at the Conference on Computer Vision and Pattern Recognition (CVPR), an annual scientific conference organized by IEEE. The AI model used, called Stable Diffusion, consists of 1.5 billion parameters and allows precise control over image generation.
One of the advantages of this approach is that it does not require an internet connection or cloud service, as all computations are performed locally on Qualcomm’s Snapdragon 8 Gen 2 chip. This makes the process more cost-effective, efficient, and offers better customization, data protection, and security. Additionally, this method helps reduce the electricity consumption associated with AI image generation, which is an important concern in the context of climate change and energy reduction.
However, it’s worth noting that this demonstration is just a first step and the results achieved are still far from the performance of more advanced AI models. Google is also interested in this approach and aims to enable AI-based image generation directly in Chrome. Meta, on the other hand, has developed a language model called LLaMA, which can operate on a computer and opens new possibilities in the field of text generation.
In conclusion, local image generation on smartphones in just 12 seconds represents a promising advancement in the field of AI. The efforts of Qualcomm and other major industry players to develop AI models that operate on mobile devices offer new possibilities in terms of performance, customization, and security while contributing to reducing energy consumption.