Meta presented CM3leon: and’Generative Artificial Intelligence (AI). which allows you to create images from a text prompt, but also to describe images using text. And Meta promises it will be the most advanced of its kind.
CM3leon, the new generative AI to create Meta images
Meta has decided to challenge the many different artificial intelligences with “graphic” capabilities that we have seen being born in recent months. As Stable Diffusion, DALL-E, Midjourney e Bing Image Creator. And it does so by promising that its technology it is the most efficient of all.
It’s a text-to-image technology, but it’s different from the “broadcast” tools we’ve seen so far. According to Meta, “Diffusion models have recently dominated imaging activities due to the their performance and relatively modest computational costs. In contrast, token-based autoregressive models are known to produce results with excellent performance and can offer even better overall image consistency“.
In other words, harder to train them for Meta. But the results for users promise to be better.
Instead of training the model by scraping images available online, a practice that raised copyright concerns, used licensed images from Shutterstock. They then trained the tool in a “supervised” manner, analyzing the results created with human experts.
In the Meta post, the company announced the high-quality results in the image generation and editing on text inputs (for example by modifying paintings and images). But he can also write image captions and much more.
At the moment, however, this is an internal study: we don’t know if Meta will launch the tool on the market. We will keep you posted.