What's this about?
A unified multimodal model that calls tools during image generation — generating images "agentically," in steps that can invoke external capabilities, rather than producing them in a single forward pass (13↑).
What exactly is "agentic image generation"?
Instead of one-shot "prompt in, image out," the model works like an agent: it can take intermediate actions — call a tool, check a result, refine — while building the image. Think of an artist who can reach for reference material and tools mid-drawing rather than painting blind in one stroke.
Why does it matter?
It merges two hot threads — agents that use tools, and unified models that both understand and generate. Agentic generation is plausibly how you get more controllable, correctable images (vs. the roll-the-dice quality of one-shot generation), relevant to the week's creative-build demos.