How to Write a Text-to-Video Prompt That Holds Together

Text-to-video prompts fail in a predictable way: they describe a story instead of a shot. A story has characters, plot, and multiple locations. A shot has a subject, an action, a camera, light, and atmosphere. The model can generate a shot. It struggles with a story.

This guide gives you a five-line structure that works for short clips, plus the review process that catches the failures.

The short answer

Write five lines: subject, action, camera, light, atmosphere. Keep the action to one thing and the camera to one move. Choose the aspect ratio before you write. Start with the shortest clip the model supports. Inspect faces, hands, text, and motion frame by frame.

Line 1: the subject

Name one main subject. It can be a person, an object, an animal, or a place.

A lone hiker in a red jacket.

If you include a person, describe them in a way that does not require a specific real identity:

A middle-aged hiker in a red jacket and a dark backpack.

Avoid naming a real actor or a public figure. The model may produce a likeness you do not have the rights to use.

Line 2: the action

One action. Not a sequence.

Walks slowly along a ridge.

Good actions for short clips:

• Turns their head.
• Walks a few steps.
• Raises a hand.
• Looks out at the horizon.
• Opens a box.
• Pours a drink.
• Rotates slowly.

Actions that are hard:

• Fighting.
• Dancing with complex footwork.
• Handing an object to another person.
• Running and turning.
• Anything requiring precise hand contact.

If you need a complex action, split it into multiple clips and edit them together.

Line 3: the camera

One move, with a speed and a direction.

Slow push-in from a wide shot to a medium shot.

Camera vocabulary that works:

• Static camera.
• Slow push-in.
• Slow pull-out.
• Lateral pan.
• Low-angle shot.
• Over-the-shoulder shot.
• Drone-style rise.

Combine at most two ideas, such as "slow push-in, slightly low angle." A list of five camera moves will confuse the model.

Line 4: the light

Light determines the mood and the color.

Soft sunrise light from the right, warm highlights, cool shadows, light haze.

Useful light descriptions:

• Soft window light.
• Hard noon sun.
• Overcast diffused light.
• Neon night light.
• Golden hour backlight.
• Single softbox.
• Candlelight.
• Blue hour.

Name the direction and the quality, not just the color.

Line 5: the atmosphere

Atmosphere adds depth and motion without new subjects.

Thin fog drifts between the ridges, a few birds in the distance.

Useful atmosphere:

• Fog or mist.
• Rain or snow.
• Dust motes.
• Smoke.
• Light shimmer.
• Heat haze.
• Waves.
• Leaves in the wind.

Keep the atmosphere away from the face if identity matters.

Put it together

A lone hiker in a red jacket walks slowly along a mountain ridge.
Camera: slow push-in from a wide shot to a medium shot, slightly low angle.
Light: soft sunrise from the right, warm highlights, cool shadows.
Atmosphere: thin fog drifts between the ridges, a few birds in the distance.
Cinematic, photorealistic, stable anatomy, no text, no logos, 5 seconds.

That is a shot. It has enough detail to guide the model and not so much that the model has to improvise a plot.

Aspect ratio and duration

Choose the ratio before you write. A vertical clip needs a different composition from a horizontal one.

Keep the first clip short. Short clips are cheaper, easier to inspect, and easier to keep coherent. Extend only after the motion is stable.

Inspect the result

Pause on frames. Check:

• Anatomy: fingers, hands, arms, legs, and faces.
• Identity: does the person stay the same?
• Text: signs, labels, and screens.
• Physics: do objects move with weight?
• Background: does it warp, repeat, or disappear?
• Continuity: does the subject stay in the same place?
• Ending: does the last frame hold together?

A clip can look fine at play speed and fail on a single paused frame. If the frame fails, the clip fails.

Common mistakes

Writing a story. The model needs a shot, not a plot.

Adding multiple characters. Each character adds a chance for anatomy and identity failures.

Asking for text on a sign. Text is unreliable. Add it in an editor.

Using a complex action. Split it into simpler clips.

Ignoring the ending. The last frames often drift.

Forgetting the aspect ratio. A horizontal prompt cropped to vertical loses the composition.

Assuming audio. Audio depends on the selected model. Check the live controls.

When to use image-to-video instead

Use image-to-video when you already have a strong still and want to add motion. It gives the model a fixed first frame, which usually improves identity consistency.

Use text-to-video when you need a new scene and do not have a source image. It has more freedom, which is both the advantage and the risk.

If you have a good still, read the image-to-video guide first. It is often the more reliable path.

A reusable video prompt template

[One subject with a clear visual description].
[One action].
Camera: [one move, speed, angle].
Light: [direction, quality, color].
Atmosphere: [one atmospheric element].
[Style: cinematic / documentary / commercial].
Stable anatomy, consistent identity, no text, no logos.
[Duration and aspect ratio supported by the model].

Try the workflow

The AI video generator page opens the workspace in text-to-video mode with three shot prompts. You can copy a prompt without an account. A generation uses credits; the pricing page lists the current packs and plans.

If you need a thumbnail for the finished clip, read the thumbnail guide. If you want to turn a still into motion, start with the image-to-video guide.