Meta has launched its Muse Image model for image generation and previewed Muse Video, enhancing media content creation capabilities.
Meta has officially launched its Muse Image model for generating images and has offered a sneak peek at its upcoming Muse Video feature.
Developed by Meta's Superintelligence Labs, Muse Image is now accessible via the Meta AI app, on meta.ai, within Instagram Stories for U.S. users, and in WhatsApp in select regions. The company touts Muse Image as its most sophisticated image generation model to date, capable of closely adhering to prompts, editing images, crafting scenes from various references, and incorporating social context from Instagram.
Muse Image functions as an agent-based model, employing internet searches and coding tools while enhancing images throughout the generation process. It integrates with Muse Spark, allowing users to create animated GIFs, websites with embedded images, and interactive visual games.
In addition to Muse Image, Meta provided an early glimpse of Muse Video. This forthcoming model is said to share the same foundational pre-training as Muse Image, promising high visual quality and native audio support. However, Meta is still working on improving audio-video synchronization and accurately depicting rapid motion.
Meta has also introduced a feature called Content Seal, an invisible watermark for images generated with Muse Image in the Meta AI app. This watermark remains intact even after cropping, resizing, or screenshotting, with plans to extend it to videos in the future.
Furthermore, Meta released performance metrics for its tools, noting that Muse Image ranks second in text-to-image generation, single-image editing, and multi-image editing, trailing only behind GPT Image 2. Earlier this year, Meta also revealed an AI system capable of reconstructing typed text from brain activity, achieving an average word-level accuracy of 61%.