Meta Launches Muse Image, Previews Muse Video with Native Audio from Superintelligence Labs
AI video model releases
Meta Launches Muse Image, Previews Muse Video with Native Audio from Superintelligence Labs
Meta Superintelligence Labs launched Muse Image, an agentic image generation model, and previewed Muse Video, built on the same pretraining base with native audio support
Muse Image is available now in the Meta AI app, meta.ai, Instagram Stories in the US, and WhatsApp in limited countries, with Facebook support coming soon; Muse Video is coming soon to creators and Meta AI
Instead of directly mapping prompts to images, Muse Image acts as an agent that invokes coding and search tools, self-refines its own generations during chain of thought, and scales test-time compute for better accuracy
During reinforcement learning, Muse Image learns to write and execute code for accurate plots and QR codes, and integrates with Muse Spark to create animated GIFs, websites with embedded images, and interactive visual games
Self-refinement behavior in Muse Image, including local edits or full regenerations when parts of an image are wrong, emerged during RL training without explicit design because it produced higher reward
cho.sh