Recently, Meta has launched a new generation of AI-powered image creation tool called “Imagine with Meta AI.”
According to a report published on arstechnica, approximately 1.1 billion publicly shared images on Facebook and Instagram were utilized to train the new artificial intelligence model.
It is evident that this approach raises some concerns about user privacy. If you don’t have an account on Facebook or Instagram, or if your account is closed, they may have used your images to train “Emu.” It is known that Instagram users have been uploading over 95 million images daily since 2016, meaning that the data used to train the current model may be much less than expected.
“Imagine with Meta AI” relies on the artificial intelligence model “Emu,” which is the foundation for the new image creation features. Meta published a research paper in September explaining that Emu is capable of generating high-quality images through a process known as “quality tuning.”
According to the research study, Emu was trained on 1.1 billion pairs of images and texts, then fine-tuned using very few high-quality images, resulting in an 82.9% success rate compared to other models.
Meta did not clearly specify the source of the training data, but reports indicate that Nick Clegg, Meta’s head of global affairs, mentioned the use of social media posts as a source of training data, including images fed into Emu.