I want to use AI on my phone, but I'm stuck in a dead zone. Usually, these large language models require an internet connection, so it won't work. Ah, but wait. Google cooked an app designed to overcome that barrier.
Now available on Android and coming to iOS soon, the experimental Google AI Edge Gallery will download the necessary models and files you need to use AI. You can then perform normal tasks, such as asking questions and finding information, even when offline.
Also: Google Gemini Assistant is getting 8 useful features – this is the update log
“The Google AI Edge Gallery is an experimental app that takes advantage of the power of cutting-edge generator AI models directly, running perfectly on Android (available now) and iOS (coming soon) devices,” Google said on the app's GitHub page. “Once your models are loaded, you'll dive into a world of creative and practical AI use cases where everything runs locally, without the need for an internet connection. Try out different models, chat, ask questions with images, explore prompts.
How does this work?
The setup process is tedious, but feasible. Because it is experimental, the Google AI Edge Gallery is not available on Google Play or other official app stores. Instead, you need to install it via an APK file. To do this, you will need to adjust the settings on your Android device to allow installation of unknown apps. Additionally, you will need to have a face hugging account and download the LLM version running locally.
“On your device, go to Settings and select App Settings. Depending on your phone and OS version, tap the options at the bottom for “Special App Access” or tap the 3-dot icon at the top and select “Special Access.” On the next screen, tap the “Install unknown apps” option and turn on the Chrome you want to allow from that source.
Also: How to try Google's AI Video Generators Going Viral on the Internet Veo 3
To grab the app, go to the GitHub page via Chrome on your Android device. Swipe past the screenshot, then tap on the link to “Download app: Grab the latest APK.” If you receive a notification that your file may be harmful, tap the Download option anyway.
Tap the link that opens and tap Install. After installation, open a tap to launch the app. You can also find the app icon in your app library for future use.
To get started, the app offers three example tasks to try – Images, AI Chat, Prompt Lab. Tap one of them. You will then need to select one or more LLMs to download and use. Here, the app offers various versions of Google's Gemma LLM. It's a lightweight and open model similar to the Gemini, but can be run locally. Tap the double arrow next to any version to see more details. Otherwise, select one and tap the “Download and try it.”
You must create or sign in a hugging face account to continue. After logging in, you must agree to the license terms. The process can be clumsy, but don't give up. The selected version will be downloaded and installed on your device. Some of them are in a few gigabytes in size, so be patient. You can then download other versions if you wish.
Also: 3 ways to change how Google's AI mode shop online – if it works
When you're done, go back to the main Google AI Edge Gallery screen and tap on one of the three examples. Select the downloaded model to use,[トライ]Tap. For example, select AI chat and then enter or speak the prompt. In response, the AI generates that response. We asked when and where pizza was invented, but the AI provided a detailed response.
Next, try Ask Image. Here you can upload images from your device or take a new photo before asking questions. However, when I tried this I came across some hiccups.
My experience
I took a photo of the USS Enterprise model from the original Star Trek and asked how many crew members were involved in the AI. The AI correctly identifies the ship, but continued to think of very inaccurate numbers of crews, such as 22 and 10 (the actual number is about 430). After I called it with each mistake, I apologized and said it was still in development and still learning. I tried different models, but it kept giving me the wrong answer.
Also: How to run local LLM as a browser-based AI with this free extension
I also asked them to show me the cover photo of Amazing Spider-Man #1 and identify the comic book. On my first attempt, it informed me that it was an amazing Spider-Man #2. When I said the answer was wrong, it provided the correct information.
Beyond hallucinations, the app itself crashed several times while I was using the Ask Image prompt. Of course, the AI Edge Gallery is labelled as an experiment for reasons. Don't expect stable performance or accurate answers in full. But if Google can tweak the app, it can pave the way for an AIS that can run effectively, whether online or not.
Want more stories about AI? Sign up for Innovationour weekly newsletter.
