Here’s a Closer Look at Meta’s New Standalone AI App

Meta launched a standalone app for its Meta AI assistant on Tuesday, marking a significant step in bringing its generative AI tools, powered by the open-source Llama AI models, to users beyond Facebook, Instagram, and WhatsApp.

The new Meta AI app arrives nearly two years after the company began its major AI integration push in December 2023, embedding generative AI into its social media and communications platforms.

Meta is playing catch-up to OpenAI, Google, and Elon Musk’s xAI, and others who have already launched standalone apps for their AI models.

Despite having 7.5 billion monthly active users across its social media platforms, Meta faces a significant challenge if it wants to match its rivals. ChatGPT alone had 600 million monthly active users across iOS, Android, and Desktop applications.

I spent some time with the new Meta AI app, available on iOS, Android, and the web. It offers a handful of good features, like high-definition image and video generation, customizable AI voices, and a Discovery tab to see how others are using it.

There’s also integration with Meta’s smart glasses. However, even with these highlights, the app still has some rough spots that need improvement.

First look

For this review, I used the iOS version of the Meta AI app. It requires users to sign in with a Facebook or Instagram account, or with an email address tied to other Meta accounts. Once logged in, the landing page offers tutorials on using AI, exploring features, and launching simple projects.

A new feature of Meta AI is the Discovery tab, which shows what other users are creating with the app. The Discovery tab is more than a gallery—it also serves as a social media hub, allowing users to like, comment on, and share AI-generated content across Meta platforms.

While I don’t mind sharing my creations online, I didn’t like how Meta AI shares images, as it does so without first obtaining the user’s permission. Users can turn off sharing to Facebook and Instagram in their settings under Data & Privacy, but it is turned on by default.

Meta AI can create both still images and short video clips, which is one of my favorite aspects of Meta AI. Meta’s Emu image generator is very good at creating vibrant, high-definition images that can be animated into clips by adding the word “video” to the prompt. Image generation is fast.

I was surprised to find that the AI had no problem creating images of Superman, Wonder Woman, Batman, and John Constantine without much fuss.

Once created, these images can be shared along with their prompts in the Discovery tab feed.

Meta AI’s controls also include a history tab that shows previously used prompts. Additionally, under Data & Privacy, users can make their prompts private or delete them in the “Manage your information” section.

The Meta AI app also connects to Meta’s line of AI-powered glasses, indicated by a glasses icon at the bottom of the screen. For Meta Glasses owners, the app provides step-by-step instructions for pairing the device.

“You’re going to be able to use the app to manage your Meta glasses and other kinds of AI devices we’re going to be building in the future,” Meta CEO Mark Zuckerberg said in a Facebook post. “This is the beginning of a long journey to build this out.”

I’ll have to take Zuckerberg’s word for it. I don’t own a pair of Meta glasses, which start at $299, but if Meta would like to loan me a pair, I would be happy to review its integration with Meta AI.

Voice controls

While Meta AI, on the surface, is like every other large language model, what makes the app interesting and creepy is its voice AI feature.

Once enabled, the Ready to Talk feature allows the mobile app to listen for commands as soon as it opens. Which is fine until you realize that the app doesn’t stop listening after it answers your question. I had to tell the AI to stop before the microphone feature would turn off.

Voice controls are accessible next to the user’s profile picture. Users can switch between the default AI voice and celebrity voices, including Judi Dench, Kristen Bell, John Cena, Keegan-Michael Key, and Awkwafina.

It was creepy when Meta first rolled out the feature in September 2024, and it’s still creepy now, though having Veronica Mars’ Kristen Bell as my AI assistant is pretty cool.

The “Ready to talk” feature opens the microphone when the app launches. An experimental full-duplex mode enables more natural, real-time conversations, though it has limitations, including outdated knowledge and limited access to certain features.

Having the Meta AI app ready to listen is a double-edged sword. On one hand, it’s very convenient to be able to start talking immediately, while on the other hand, if you forget the setting is on and start speaking, the AI may assume you’re talking to the app.

Because the microphone quality on mobile phones differs by generation, whether your app can hear you clearly will depend on how many random pictures or conversations it generates, making you think you are speaking to it when you are not.

Memory

While AI models are becoming more personal and remembering what users tell them, like ChatGPT, Meta AI does not “remember” conversations by default.

However, this can be changed by instructing the AI to “remember this,” which will then add the memory to the app’s settings. Stored memories can be viewed in the Memory settings. Meta has not detailed how long user prompts are stored or how they may be used to train future models.

Final thoughts

The standalone app offers a more intuitive experience than Meta’s previous AI integrations in its social media platforms and functions similarly to mobile AI applications like ChatGPT, Grok, and Gemini.

Once engaged, Meta AI is responsive, answering questions within seconds of asking.

Meta AI’s app is very impressive for a free and open-source model, but it’s not something I would use daily. This is an app for people who spend a lot of time in Meta’s ecosystem, but I don’t see anyone trying to do more than have casual conversations right now.

It makes sense for dedicated Meta users, though it may frustrate those who prefer more flexibility and privacy.

Edited by Josh Quittner and Sebastian Sinclair