HelixWorld WASD camera moves as prompt words for Omni Flash

Sume does not accept camera key input or a camera trajectory. The nearest equivalent to HelixWorld's WASD camera control is to write the move in the prompt, for example slow dolly forward for W, and test it with a cheap 3-second 360p draft on Gemini Omni Flash 1.1, which costs $0.1125 before rounding. The mapping below is prompt wording to try, not a guarantee that the model will follow it.

From keys to words

HelixWorld's README describes actions W, A, S, D, left, right, up, down and stop, which can be combined with a plus sign. Its paper describes continuous 6-DoF camera poses injected into the model. A text prompt carries none of that, so every move below is a phrase.

A cheap test loop

Send the start of your scene as frame_images with first_frame, which is the same idea as HelixWorld's first-frame input, and put one move in the prompt. Test it at 360p first. At Sume's rate of $0.0375 a second, a 3-second draft is $0.1125 unrounded; a job is billed ceiled to the cent.

{
  "model": "gemini-omni-flash-1.1",
  "prompt": "Slow dolly forward through the doorway, the camera moves toward the window",
  "duration": 3,
  "resolution": "360p",
  "frame_images": [
    {"type": "image_url", "image_url": {"url": "https://example.com/hall.jpg"}, "frame_type": "first_frame"}
  ]
}

Limits

Keep one move per clip. Combined moves, or a sound that should pan with the camera, are where HelixWorld's design (camera pose in, stereo panning out) is ahead of a text prompt. If the draft follows the move, rerun at 720p or higher with a different Idempotency-Key.