-compute - s per s of video--
Connecting...
Your holds play back in the next 5 s window, on screen in -frames late 0 held by sync 0 page late 0 / skipped 0 buffer - link - in flight - dropped for network 0 key to screen - average compute -
World modelHy-Nomas is Tencent Hunyuan's open video world model. Describe a world, then move through it with W A S D and the arrow keys, or type an action for the subject. The TPU makes each 5 s of video while the one before it plays.
How this works

The model makes video in windows of 5 s (120 frames at 24 fps). While one window plays, the chips make the next one, so the keys you hold are recorded frame by frame and replayed in the window after; the strip under the picture shows when that window reaches your screen. Compute is the chips' time per second of video: under 1.00 keeps up with real time. An action you type is timed inside the next window and the rest of that window keeps the world's base action. Hide UI (H) leaves only the picture.

World

Start a ready world
Or make your own

Action

starts at1.0 s
lasts1.0 s

Timed inside the next 5 s window; the rest of the window keeps the world's base action. An action, preset or typed, is ready 0.8 to 1.5 s after you inject it and on screen 5.6 to 7.0 s after (TPU v6e-4, all prompt sizes compiled at start).

    Controls

    W A S D move, arrow keys turn and climb/dive. Hold to keep going, let go to stop: each window replays your holds frame by frame, one window later.

    Camera
    turn rate25
    move speed1.00x
    rotation scale2.5
    pitch limit60°
    Picture
    Preset
    Decoder
    Sync