Meta said on the 7th (local time) that it released its first image generation model, "Muse Image," developed by its superintelligence lab (MSL). Like competing products previously introduced by OpenAI and Google, Muse Image turns users' everyday language prompts into photos or drawings and can also edit existing images.
Meta said the model can be used for free on its social networking services (SNS), including Instagram and WhatsApp, through the chatbot "MetaAI," and that a subscription option will be available if usage is high. It also added an SNS-specific feature that generates images reflecting public photos by tagging a specific Instagram account. However, Meta also provided a setting that allows users who do not want this to refuse such tags.
Meta signaled it will expand adoption to Facebook, Messenger and Advantage Plus, and make it available to advertisers as well. To that end, it offers a feature to edit images by adding sketches or annotations on photos, and a feature to take an indoor photo and virtually fill the interior with products for sale.
A Meta representative told Bloomberg that all outputs made with Muse Image contain an invisible watermark and that safety measures are in place to prevent the creation of content that violates the terms, such as child sexual abuse material (CSAM). However, Meta Image is still available first only in some countries, with plans to expand the list over time.
Until early last year, Meta mainly developed open models such as LLaMA, but in the second half it set a goal of developing superintelligence, recruited Chief Artificial Intelligence Officer (CAIO) Alexander Wang, and put MSL under his leadership. Meta added that following this Muse Image, it is also developing Muse Video, which generates videos.