Version 1.2 fixes the "double-jointed horror" of v1.1, but more importantly, it introduces a subtle awareness of sound . I know that sounds insane—it's an image model—but the poses generated in v1.2 feel like the characters are listening. Heads are cocked slightly. Hands are raised to shush. These are ghost hunters holding their breath.

If you download this model tonight (and you should, if you have the VRAM to handle its native 768px resolution), here is your homework.

The game loop in version 1.2 is rigidly split into two core phases across a day-and-night cycle, demanding precise time management and strategic forward-planning.