Applies to: Users generating video from a text prompt only (no starting image). T2V means text-to-video.
Use these six tips to write text-to-video prompts that video models follow accurately.
Put the important details first. Video models weight the start of a prompt most heavily. Lead with subject and motion; put style details last.
Use concrete, filmable language. "Golden hour backlight through dust" works better than "beautiful atmosphere."
Name the shot type: extreme close-up, medium shot, wide establishing shot, over-the-shoulder, aerial, POV (point-of-view).
Use real cinematography terms — video models respond well to them: "35mm film look," "handheld documentary style," "slow motion," "shallow depth of field" (only the subject is in focus, background is blurry).
Describe actions in time order: "She reaches for the door, hesitates, then pushes it open."
Add physics cues for realism: "hair moving in the wind," "steam rising," "fabric swaying."