How to Write Text-to-Video Prompts and Generate a Clip

Updated November 2, 2026

A vague prompt such as “a person walking” leaves the subject, location, and camera movement open to interpretation. Specify those details before generating a clip, then revise the prompt based on what the first result gets wrong.

Build a prompt from four details

Describe one clear shot using these elements:

• Character or subject: Identify who or what appears, including a few distinctive visual details.
• Action: Describe one observable action, such as opening a door or lifting a product.
• Environment: Set the location, time of day, and lighting.
• Camera movement: Choose a simple movement, such as a slow push-in, a sideways tracking shot, or a locked-off shot.

Use this template:

A [character or subject] [performs an action] in [environment and lighting]. The camera [movement and framing]. [Visual style or mood].

For example:

A cyclist in a yellow raincoat walks a silver bicycle along a quiet city street at dawn. Wet pavement reflects soft pink light. The camera tracks slowly alongside the cyclist in a medium shot. Realistic, calm atmosphere.

Keep the action manageable for a short clip. If you need several events or locations, generate separate shots rather than asking for a complete story in one prompt.

Generate a clip in Oranzy

1. Open Oranzy video generation and sign up for 100 bonus tokens if you need an account.
2. Enter your prompt, including the subject, action, environment, and camera movement. Start with one shot and one main action.
3. Select an available video model, such as Kling 4.0, Seedance 2.5, or Wan 3.0. Choose an aspect ratio suited to the destination: 9:16 for vertical social video, 16:9 for landscape video, or 1:1 for a square placement. Check the options and token cost shown before generating.
4. Generate the clip and review it at the intended display size. Check whether the action is clear, the subject stays recognizable, and the camera moves as requested.

Model availability and settings can vary, so use the choices shown in the video tool. If your concept starts with a product image rather than text alone, see the guide to turning a product photo into a sales video.

Refine the result with a second prompt

Change the detail responsible for the problem instead of rewriting everything. If the cyclist looks correct but the camera moves too quickly, keep the subject and setting and make the camera instruction more specific:

A cyclist in a yellow raincoat walks a silver bicycle along a quiet city street at dawn. Wet pavement reflects soft pink light. The camera moves slowly alongside the cyclist at walking speed, maintaining a steady medium shot. Realistic, calm atmosphere.

Generate again and compare the two clips. If the action is still unclear, simplify it further: ask the cyclist to walk beside the bicycle rather than walk, turn, and mount it in the same shot. Save the prompts that produce useful results so you can adapt them for later clips.

Avoid common mistakes and wasted generations

• Overloading the shot: Multiple characters, actions, and camera moves can compete. Start with one subject and one action.
• Using conflicting instructions: “Locked-off camera” and “camera circles the subject” describe different shots. Pick one movement per prompt.
• Leaving out visual context: Add a location and lighting direction or quality when they matter to the result.
• Choosing the wrong frame: Decide between vertical, landscape, and square before generating. Important details near the edges may be lost if you crop later.
• Changing too much at once: Revise one major element per attempt so you can tell what improved the clip.
• Upscaling a flawed take: First select a clip with the right action and composition. Then consider AI video upscaling if you need a larger output; upscaling will not fix an incorrect action.