Muse Image & Muse Video

The first image-generation model from Meta Superintelligence Labs, with a video sibling in preview: agentic generation with tool use during synthesis, and a "Content Seal" provenance watermark. At launch, Arena #2 in text-to-image and #3 in text-to-video. Proprietary; parameters and training details undisclosed. Note: the companion Instagram photo-editing feature was pulled within days after user backlash, but the models themselves remain deployed.