What capability makes this possible? It’s incredible! Does it have an image-generation model?

No this is just it moving the mouse and viewing the results with it's vision capabilities.