Meta has officially entered the AI media generation arena. On July 7, 2026, the company’s Superintelligence Labs (MSL), led by Alexandr Wang, unveiled Muse Image — its most advanced text-to-image model — alongside a preview of Muse Video, a text-to-video model with native audio support. This marks Meta’s first in-house media generation models, replacing its previous reliance on third-party tools like Midjourney and Black Forest Labs.
Agentic by Design
What sets Muse Image apart from conventional image generators is its agentic architecture. Rather than mapping prompts directly to pixels, the model functions as an agent: it can invoke search and coding tools to improve factual accuracy, write and execute code for precise charts and QR codes, and self-refine its own generations through chain-of-thought reasoning. This self-refinement behavior wasn’t explicitly programmed — it emerged during reinforcement learning because producing better images yielded higher reward.
Muse Image also integrates with Muse Spark, Meta’s flagship reasoning model, allowing the two models to share tools and plan jointly for complex agentic media generation workflows — from animated GIFs to interactive visual games.
Benchmark Performance
Muse Image claims the #2 spot on the Arena leaderboard for text-to-image, single-image editing, and multi-image editing as of July 5, 2026. Muse Video ranks #3 for text-to-video. Internal benchmarks show Muse Image outperforming Google’s Nano Banana 2 on editing tasks while trailing OpenAI’s GPT Image 2 — a respectable position for a first-generation model.
Deep Product Integration and Monetization
Muse Image is available today across Meta AI (app and web), Instagram Stories in the US, and WhatsApp in select countries, with Facebook and Messenger coming later this year. Meta is pairing the launch with a clear monetization strategy: free tier access for casual users, with power users routed toward the Meta One subscription plans introduced in May. Advertisers gain access to Muse Image through Advantage Plus, Meta’s AI-powered ad creative platform.
All Muse Image outputs carry Content Seal, Meta’s invisible watermarking system that survives cropping, compression, and screenshots — a nod to the growing demand for AI content provenance.
Strategic Significance
This launch signals Meta’s ambition to vertically integrate its AI stack and reduce dependency on external model providers. Coming just three months after Muse Spark replaced the Llama family as Meta’s flagship language model, Muse Image and Video complete the multimodal picture. The message is clear: Meta Superintelligence Labs is building a full-suite AI platform, and it’s moving fast.