A cinematic result can still change the captured scene. In this example, Gemini Omni Flash takes the recorded video and text prompts but, without the target point-cloud video, replaces the blue board and captured room with a yellow board in a different room. The point-cloud condition specifies the planned view and timing and helps preserve the captured scene.

Gemini Omni Flash with point-cloud video

The planned camera move keeps the blue board and the captured room.

Gemini Omni Flash without point-cloud video

Recorded video and text prompts alone. The result changes the board, room, and lighting.

Recorded handheld input

The phone passes over the board because it cannot fit through the opening.

Target point-cloud video

The offline Pi3X proxy rendered along the planned path, with the camera's timing already applied.