13/12/2023
December 13, 2023
Meta launches "Imagine with Meta" in the U.S., a free AI tool generating photorealistic images from text prompts. Powered by Emu, it rivals platforms like DALL-E. Features include downloading images and an upcoming invisible watermark for AI-generated content.
Meta has recently launched "Imagine with Meta," an exclusive platform in the United States enabling users to create photorealistic images using a simple prompt.
Developed by Meta, this free-to-use tool swiftly generates four photorealistic images in a matter of seconds. "Imagine with Meta" marks Meta's entry into the burgeoning field of AI image generation, currently dominated by players like Midjourney and DALL-E.
Emu, Meta's internal image generation model unveiled a few weeks ago, powers "Imagine with Meta." The platform is standalone and can be accessed freely at Meta's website. Users need a Meta account to use it. However, its availability is currently restricted to the U.S., with no confirmed release date for France.
"Imagine with Meta" operates on a text-to-image basis, similar to other platforms in the market. Users can enter a description of their desired image, and the tool produces four photorealistic images in JPEG format, which can be downloaded. According to Ars Technica, Emu is trained on over 1.1 billion images from Meta's social platforms like Facebook and Instagram.
It's important to note that this feature isn't entirely new. Previously, "Imagine" was part of Meta AI, an AI assistant revealed at Meta Connect. This chatbot, still in beta in the U.S., functions similarly to ChatGPT or Google Bard, handling requests, offering recommendations, and generating photorealistic images. Meta's decision to create an independent experience with "Imagine" was inspired by the beta version's success:
"We've been thrilled to hear how people have been using 'Imagine', the image generation feature of Meta AI, to create fun and creative content in conversations."
For transparency and traceability, Meta announces the integration of an invisible watermark in the coming weeks. This watermark, undetectable to the human eye, can be identified using a corresponding model. It withstands image manipulations like cropping, color changes, and screenshots. However, each photorealistic image generated by the tool already has a visible watermark in the lower-left corner.
