Header ads

Qualcomm Shows Off Fastest-ever Local AI Image Generation on Mobile -Gamekhol

The stable diffusion of a wide range of AI models pressed onto smartphones is the next step in bringing generative AI tools to a wider audience. Apple released its stable Diffusion Optimization with iOS model last December.

Qualcomm is making the best of its AI chips on mobile. It is claimed to be the fastest and best deployment of AI image generator stable diffusion ever on a smartphone.

A demo video showed Qualcomm generating a 512 x 512 pixel image within a few seconds. It shows version 1.5 of a stable diffusion. Although Qualcomm will say that it doesn't know the mobile, it says that it manages the flagship Snapdragon 8 Gen 2 chipset inside it more smoothly. (Which was launched last month and is powered by Hexagon processors with AI at the core). An important stable diffusion has been prepared by the engineers of this company. All kinds of custom optimization done on one software for how to run it.
ai image stable diffusion mobile android qualcomm fastest

In some contexts: running a program like stable diffusion (which is a first in AI image generation) requires a lot of computing power. For that many applications that provide such services on mobile phones burn your mobile. Instead, all that processing takes place in the form of the cloud. Even if you want to generate the image on a tablet or laptop in this way, it will take a few minutes. Getting a 512 x 512 picture every second from the phone is impressive.

Qualcomm claims that its speed is the fastest which is a record. We have no reason to doubt it either. The company's report said that it was the first time that Stable Diffusion was installed natively on Android, which no one thinks is true. After some research I found this blog post written by a developer Yvonne Huang showing how they are running Stable Diffusion on Sony Xperia 5 II smartphone powered by Qualcomm Snapdragon 865 and 8GB RAM. A tweet by Huang revealed that it took some time to generate a 512 x 512 image with this setup. Qualcomm has certainly brought its 'speed' to the fore, not to be technically "first".

Another is when compared to iOS. In December, Apple saw its machine learning framework core ML running natively. Several necessary optimizations were released to obtain its stable diffusion. Today we found Stable Diffusion 1.5 on the iPhone 13 powered by the Draw Things app with Core ML acceleration to test that system extensively. If done this way, it took about 60 seconds to generate a 512 x 512 image. So again Qualcomm's speed was the best sample. Although several caveats apply, Qualcomm is using existing hardware and custom optimization packages. which has not been made publicly available. While iOS was tested third-party apps on 2021 smartphones.

On all these sides, this Qualcomm is also impressive, it's just a demo. There are all the major local tools on the base of Pruitt Comp, from mobile powered models. There's convenience (doesn't require a mobile connection), cost (developers don't charge users when the server is billed), and privacy (running locally means you send data to another computer).

Post a Comment