Meta unveiled a novel complementary stand-alone AI image generation platform called Imagine with AI on Wednesday. This image-generative tool is based on its Emu image synthesis model, and its training data contains 1.1 billion publicly accessible images on Facebook and Instagram. With this capability, you can easily generate unique images based on the textual prompts.
This is the era of generative AI, and every day new updates are rolling out on the internet. Today, it is the turn of Meta, and it is taking the spotlight by introducing AI tools across its services and on the website. The company has launched a new AI image generator tool that is trained on a wealth of Facebook and Instagram photos.
Before, this technology was limited to Meta’s messaging and social networking apps like Instagram. Now, it is widely available on the website.
If you are already on Facebook or Instagram, then there is a chance that your images present on social media also contribute to the Emu training. As you have heard the well-known saying, ‘‘If you’re not paying for it, then you are the product.’’
Back in 2016, Instagram users uploaded more than 95 million photos daily, however, the data set employed by Meta to train its AI model is 1.1 billion which is a very small fraction of the overall photo library.
Meta gives its assurance that they exclusively utilizes public photos for its training purpose and that they don’t touch private photos. So, if you choose to set your photo to private on Facebook and Instagram, then your photos won’t be incorporated into the AI model training unless there is a change in the policies in the future.
Meta expressed its pleasure by receiving positive feedback from users who have shared their experience of using Imagine with AI and crafting fun and creative content in chats. In their official announcement, they declared that they are expanding access to Imagine outside of the chat, making it available in the United States through Imagine.meta.com.
This platform is specially designed for creative enthusiasts and lets them create the image by using the technology from Emu, Meta foundation image model.
Just like a Stable Diffusion, DALL-E 3 and Medijourney, the Imagine with Meta AI platform that is trained on the visual data produced novel images. To utilize this, users are required to have a meta account, which can easily be imported from the existing Instagram and Facebook accounts.
Each generated image is 1280 by 1280 pixels in JPEG format that also contains the watermark of Imagine with AI, but it can easily be covered up by users. Meta plans to introduce the invisible watermark which is a digital watermark embedded in an image data and cannot be visible to the naked eye. This watermark will be impossible to cover up compared to the current ones.
Meta model demonstrates its proficiency in generating photo-realistic images but it still falls slightly short of the exceptional quality achieved by Midjourney. It is proficient in handling intricate prompts and surpasses the DALL-E capabilities.
Overall, Meta image generation capabilities seem to align with the industry average. It offers a competitive but not groundbreaking performance in the current landscape of AI-driven image generation.
The latest evolution of AI stirs a commotion, and the obvious issue is that people who shared their art and photos online before the advent of AI will use their artwork to train the AI. Just because someone shares their photos on the internet doesn’t mean that they want their work to be used to train AI tools. This is an ongoing debate on getting permission to use art and written work for AI training among communities that also currently undergoing legal battles.