How to write AI video prompts

Start with a single visible action

A useful prompt specifies subject, action, environment, camera and light. Write what changes over time: a robot places a sphere on a table is clearer than a beautiful futuristic robot. Keep one action in a first test.

Separate the camera from the subject

State whether the subject stays still and the camera pushes in, or the camera stays fixed while the subject moves. Combining zoom, orbit and cuts can create contradictory instructions.

Iterate with a small experiment

Keep duration, orientation and the source material unchanged while changing one instruction. Compare movement, composition and unwanted details. A prompt is guidance, not a precise camera path.

Turn feedback into a revision

If the scene barely moves, name a concrete action. If faces distort, reduce turns and occlusions. If the scene changes abruptly, remove cuts and competing settings. Save the exact prompt with each downloaded result.

Settings and credit budget

Choose 5, 10 or 15 seconds. Text tasks cost 6 credits; image and two-frame tasks cost 8. New Google accounts receive a one-time 60-credit grant valid for 30 days. Landscape output is 896 × 512 and portrait output is 512 × 896 at 24 fps. Each account can have one active task. Wait for a confirmed result instead of repeatedly submitting.

从这个提示词开始

银色机器人在深色影棚桌面上小心放下玻璃球，镜头从中景缓慢推进，柔和轮廓光映在玻璃上，一个连续镜头，无切镜。

可从这个提示词开始，每次只调整一个动作或镜头指令。它是提示词示例，不是对特定成片效果的承诺。