Meta just dropped its first image model from the Superintelligence Labs, and it’s a bit of a mixed bag. Technically, Muse Image is savvy—it works like an AI agent, pulling in web searches and fixing its own mistakes during generation. This self-correction emerged naturally from reinforcement learning. It’s ranking second in human preference tests, behind OpenAI’s GPT Image 2 but ahead of others like Nano Banana. The model can also edit images precisely, keeping backgrounds consistent across changes. But there’s a catch: you can now generate images of anyone with a public Instagram profile just by tagging their username, without needing their permission. Privacy watchdogs in Europe are already circling. Why it matters: While the agent-like approach marks a leap in AI image accuracy and creativity, the consent loophole could sour trust before the technology matures.