Snapdragon 8 Gen 3 – What AI is Actually Available on the Device?


Artificial intelligence (AI) available on devices will be a hot term in the coming months, especially on devices that use the Snapdragon 8 Gen 3 chip. Many may still be unclear as to what is meant by "available" AI and what is the difference with AI that can accessible now. We were lucky enough to be able to try some of the AI demos available on devices that will be used more widely at the Snapdragon Summit.


Fast Stable Diffusion Generative AI

The most amazing demo we saw was Stable Diffusion on a smartphone with Snapdragon 8 Gen 3. According to Qualcomm AI Stable Diffusion by Stability Lab can run directly on Snapdragon 8 Gen 3 devices with over 7 billion parameters. When the prom is inserted into the generated application, the image can be generated in 0.6 seconds. This beats the 15 seconds it takes the Snapdragon 8 Gen 2 to do the same process.



All of this is done without an internet connection. The processing available on the machine will allow the owner to generate background images like the Pixel 8 Pro can do in times shorter. It is also many times faster than using an internet connection where the same program is processed in the cloud first before being sent to the device.



We've already tried some generative AI this year like Google Duet in Singapore three weeks ago and nothing is as fast as what we saw in the Snapdragon demo.


Faster Photo Editing With AI

Google Photos has a Magic Eraser feature that can precisely erase background objects. The AI used by Google for this feature uses a Tensor chip that is exclusive to Pixel devices. At the Snapdragon Summit we tested the same feature on a demo device. Not only background objects can be deleted in an instant, we can also change the objects in the picture.



In a demo at reruai, we instructed the AI to transform an elevator door in a picture into a window showing a Hawaiian beach. Like the Stable Diffusion demo above, this image is generated by Qualcomm's software in just a few seconds. Several iterations of the image can be generated by simply pressing a button in the application. Also featured is the Photo Expansion feature that generates additional images that were not in the original photo. It is very similar to the Generative Fill feature in Photoshop.



Before and after Photo Expansion

According to a Qualcomm representative, this opens the door for device owners to change the color of the sky, night to day, zoom out like an ultra-wide-angle lens and many other different scenarios they can think of in their heads.


A Smarter Virtual Assistant

Google Assistant is our favorite virtual assistant because it's the fastest, rarely gives weird information, and understands all commands given. But like other generative AIs, an internet connection is required at all times for the Assistant to work.



AI demo available Honor

With AI available, personal virtual assistants that are then trained with each user's personal data can be produced. Honor showed an AI demo that was instructed to produce videos for children through personal videos in the gallery app. The AI realizes the user has more than one child and then asks which child to choose.


The next instruction was given to choose one of the children but only a video of him dancing and smiling. Then a short video is generated by AI without the owner having to manually edit or search for video clips manually. During the demo only the virtual assistant screen with AI is displayed on the screen.


Advantage

The advantage of AI available in the device is not limited to speed and does not require an internet connection. The real advantage is privacy. With all data processed in the device without first being uploaded to the cloud, users don't have to worry about their data, photos and videos being leaked to the web arena. A virtual assistant that really understands daily routines can also be trained with personal data. Each virtual assistant with AI will be unique to each individual. It seems to realize the fictional world of HER directed by Spike Jonze a decade ago.



Today reports surfaced that Samsung wants to make the Galaxy S24 their first AI-powered smartphone with ChatGPT and Bard available. Xiaomi 14 has generative AI that generates images based on real user photos. These are just two first-generation devices, with smaller AI models paired with powerful chips that may soon usher in the Android smartphone era.

Previous Post Next Post

Contact Form