The digital universe is enriched with something new: “Imagine with Meta AI", the new AI image generator by Meta. This technology, powered by over a billion public images (yes, that's right) from Facebook and Instagram, promises to transform simple text descriptions into creative images. A tool that is not new but which will make people talk, given that privacy and digital ethics play a fundamental role.
Topics of this article:
How the AI image generator works
Meta has recently revealed "Imagine with Meta AI", One free web platform which acts as an AI image generator, using the image synthesis model Emu. This technology is based on an impressive database of 1.1 billion public images on Facebook and Instagram. The underlying idea is simple: starting from a textual description, the AI model is able to create new and original images. Ultimately it's about the shape text-to-image.
Emu stands out for its ability to generate high-quality images through a process called "quality-tuning". Unlike traditional text-image models, Emu focuses on aesthetic alignment after a pre-training phase, using a set of visually appealing images.
The peculiarity of “Imagine with Meta AI” lies in its training: public images of popular social media become the fertile ground for teaching the AI visual concepts. This raises intriguing questions about the use of personal data online: A published photo may have contributed to the development of this technology.
In this regard, Meta ensures that only public photos are used, therefore make your images private on Facebook or Instagram should exclude them from future training sessions.
In practice, Meta's AI image generator works similarly to other models such as Stable Diffusion, DALL-E3 and Midjourney. To create images with this new site, you need a Meta account, which can be imported from Facebook or Instagram. Every generation produces four 1280x1280 images pixels, saveable in JPEG format, with a small watermark “Imagined with AI".
The quality of images generated by Meta is generally realistic, though not at Midjourney's level. It handles complex prompts better than Stable Diffusion XL, but perhaps not as well as DALL-E 3. La ethnic diversity in human representations it is a strong point, but the model shows limitations in the rendering of texts and various media outputs such as watercolors or embroidery.
Despite the fun experience that Meta offers with its AI image generator, ethical and privacy concerns emerge. The question of intellectual property and ethical use of data remains open, raising important discussions in the era of advanced artificial intelligence.
When will it be available
At the time of its release i.e. December 6, 2023, Meta's AI image generator it is not available worldwide. In fact, if you proceed on official website for the first time and you create an account (connecting it to Facebook or Instagram), you will arrive at a web page that informs us that “Imagine with Meta AI” is not yet accessible.
Respect for privacy
We don't know when Meta will make the service accessible in Italy. There probably are problems related to respect for privacy, So as it was for ChatGPT in its time. In fact, it should not be legal (or at least regulated differently) in our country to use public photos to train the AI model.
