Write AI video prompts that describe motion and mood

The Video tab suggests this prompt: "A drone shot rising over a foggy coastline at sunrise." It holds most of what a clip needs. It has a camera move (rising), a subject (the coastline), a condition (fog) and a time of day (sunrise). Build your own prompts on the same frame.

Describe one shot

A clip is a single shot, so write a single shot. Cover these parts in a sentence or two:

• Camera. Name the framing and the move: wide establishing shot, close-up, insert of a detail, point of view, follow shot, slow push in, drone rise.
• Subject. Say who or what is on screen, with the details that stay the same through the shot.
• Action. Say what moves and how. "A cyclist rides toward the camera" gives the model something to animate.
• Setting and light. Place the shot: indoors, a rainy street, a beach at dusk.
• Mood. One or two words such as calm, tense or playful.

Three examples you can adapt:

• A close-up of hands pouring coffee into a white cup, steam rising, warm morning light, shallow depth of field.
• A follow shot behind a runner on a forest trail at dawn, mist between the trees, calm and quiet.
• A wide establishing shot of a night market, neon signs reflecting on wet ground, people walking through the frame.

Plan short beats and keep words out of the frame

Clips come back at a few set lengths, so plan a sequence as a string of beats. Generate the wide shot, the close-up and the detail as separate clips, then arrange them on the timeline and trim each to its best moment. Varying the framing from shot to shot keeps the sequence moving.

Video models tend to garble lettering. If a shot needs a sign, a label or a caption, keep it to a word or two, or leave it out of the prompt. Add the exact wording afterwards with a title or subtitles in the editor, where you control the text.

Steer the look

To set a style, describe what you would see: flat hand-drawn animation with clean thick outlines, handheld documentary footage, soft pastel colors. Names of shows, franchises and artists can be rejected, and visual traits give the model more to work with.

For a person, place or product that must look the same, attach a picture. Drop an image into the prompt box or type @ to pick one. A reference tells the model what something looks like, so the prompt can stay focused on what happens.

Compare takes and change one thing

Set Number of takes to x2 or x3 while you explore, then pick the best clip. Each take is a separate render and spends credits, so go back to x1 once the prompt works.

When a clip is close, open its "…" menu, choose Use this prompt, and change a single detail: the camera move, the light or the action. One change at a time shows you what each word does.

Let the assistant draft the prompt

Open the AI assistant with the Agent button and ask: "Write me a video prompt for a slow reveal of a mountain cabin." The answer appears in the chat and nothing renders. When you like it, paste it into the Video tab, or ask the assistant to generate the clip.

Write one today

Write one prompt with a camera move, a subject and an action in DepCut, render it at x2, and keep the better take.