How to Write AI Text-to-Video Prompts (With Examples)

Introduction: A Good Prompt Is the Difference Between Clips

AI text-to-video generators have turned written words into moving pictures. You type a description, and the model returns a short clip with frames, motion, and sometimes sound. Anyone can do it. Not everyone gets good results.

Run the same tool twice with different prompts and you see the gap right away. One prompt returns a flat, muddy clip that looks random. Another returns a shot that looks planned, lit, and directed. The prompt is where that difference is decided.

The good news: prompt writing is a skill you can learn. This guide covers how AI video models read your words, the building blocks of a strong prompt, a simple formula to follow, mistakes to avoid, and copy-ready examples you can adapt to your own subjects.

How AI Video Models Read Your Prompt

Text-to-video models learn by studying large amounts of video paired with written descriptions. When you enter a prompt, the model matches your words to visual patterns it saw during training.

Think of it as a translation job. The model reads "a red car driving through a desert at sunset" and breaks it into visual pieces: a car, the color red, movement, a desert, warm light. It then predicts frames that fit those pieces and stitches them into a short clip.

Not every word gets equal weight. Nouns that name the subject usually matter most. Adjectives add texture. Verbs set the motion. The order of words also matters, since models tend to give earlier words more attention.

This is why vague prompts fail. "A nice video of a beach" gives the model too many choices, so it picks one at random and you get an average clip. A prompt that names the beach, the weather, the camera, and the light gives the model a much smaller field to choose from.

The Building Blocks of a Strong Prompt

Every reliable AI video prompt draws from the same set of blocks. You do not need all of them every time, but knowing them gives you a checklist to work from.

• Subject. The main thing in the frame. Be specific: "a husky puppy" beats "a dog."
• Action. What the subject does. "running through tall grass" beats "playing."
• Setting. Where the action happens. Name the place, the weather, and the time of day.
• Camera. Angle, framing, and movement. A "slow tracking shot from a low angle" changes the whole feel of a clip.
• Lighting. Light source, quality, and mood. "golden hour light" and "soft studio light" produce very different results.
• Style. The look of the finished video. Think "cinematic," "photorealistic," "3D animation," "watercolor," or "anime."
• Technical details. Duration, aspect ratio, and motion style. For example, "16:9, 5 seconds, smooth motion."

Treat these as building blocks, not as a required list. A product clip may skip style. An abstract clip may skip subject. Choose the blocks your scene actually needs.

The Prompt Formula

Combine the building blocks into one sentence, two, or three. A working pattern looks like this:

# The formula
[subject] + [action] + [setting] + [camera] + [lighting] + [style] + [technical details]

# Filled example
a red vintage car + driving on a coastal road + at sunset + slow tracking shot + warm golden light + cinematic + 16:9, 5 seconds

Start with the core three: subject, action, and setting. That is the scene. Then add the camera and lighting to control how the scene looks. End with style and technical details to lock in the final look and format.

Read your prompt out loud when you finish. If it sounds like a scene from a script, you are on the right track. If it sounds like a wish list, cut it down.

Common Prompt Mistakes (and How to Fix Them)

Most bad clips come from a small set of repeatable mistakes. Here are the six most common, with a direct fix for each.

• Being too vague. "A video of a city" tells the model nothing specific. Fix: name the subject, the place, and the action.
• Overloading the prompt. Too many characters and objects drain quality. Fix: keep one main subject and let everything else support it.
• Ignoring the camera. Without camera direction, the model invents a static frame. Fix: tell the model where the camera sits and how it moves.
• Contradicting yourself. "A bright sunny day" plus "heavy rain" confuses the model. Fix: check that every detail works together.
• Skipping the negative prompt. Unwanted elements often appear because you never said they should not. Fix: state what you do not want in the negative prompt field.
• Giving up after one try. The first clip is a draft, not a verdict. Fix: treat it as feedback and change one thing, then run it again.

AI Text-to-Video Prompt Examples You Can Copy

These examples follow the formula above. Each one shows a prompt, then a short note on why it works. Copy them, change the subject, and use them as a base for your own clips.

1. Product Shot

# Prompt for a product video
a matte black coffee maker on a white marble counter, steam rising from the carafe, slow push-in from a low angle, soft window light, photorealistic, 16:9, 4 seconds

Why it works: one clear subject, a simple action, controlled lighting, and a single camera move. Nothing competes for the model's attention.

2. Cinematic Travel Shot

# Cinematic travel clip
a lone hiker standing on a mountain ridge at sunrise, wind moving the grass, wide aerial shot pulling back, golden light, film grain, cinematic, 16:9, 6 seconds

Why it works: the setting, time of day, and camera movement all point in one direction. "Golden light" and "film grain" set the mood before the first frame renders.

3. Corporate Explainer

# Clean 3D explainer clip
a clean 3D office scene, a light bulb turning on above a desk, camera drifting forward, bright even lighting, minimal 3D style, 16:9, 5 seconds

Why it works: "bright even lighting" and "minimal 3D style" keep the render clean, and the light bulb gives the clip a clear story beat.

4. Food and Lifestyle

# Food video for vertical formats
steam rising from a bowl of ramen on a wooden table, chopsticks lifting noodles, close-up with shallow depth of field, warm side light, photorealistic, 9:16, 5 seconds

Why it works: the vertical 9:16 format fits short-form video, and "close-up with shallow depth of field" pushes the food into focus while the background stays soft.

5. Nature Documentary

# Nature documentary shot
a cheetah sprinting across a dry savanna, dust kicking up behind it, slow-motion telephoto shot, midday light, documentary realism, 16:9, 6 seconds

Why it works: the action is singular and fast, "slow-motion" gives the model a clear frame rate to aim for, and "documentary realism" sets the tone.

6. Abstract Motion Graphic

# Abstract motion clip
layered ribbons of blue and purple light twisting through dark space, smooth camera orbit, neon glow, 3D abstract style, 16:9, 8 seconds

Why it works: no complex subject to distort, two colors, one camera move, and one clear style. Abstract prompts are forgiving because there is no "correct" form to judge.

How to Iterate and Refine a Prompt

The first run is rarely the final clip. Treat your first prompt as a draft and build from what the model returns.

1. Watch the clip and note what fails. Is the subject wrong? Is the motion odd? Is the lighting flat? Name the problem in one sentence.
2. Change one variable at a time. If you change the subject, the camera, and the style at once, you cannot tell which change fixed the issue.
3. Use negative prompts for recurring problems. If the model keeps adding text or extra hands, list them in the negative prompt field.
4. Keep a small library of prompts that worked. Small edits on a proven prompt beat starting from zero every time.

Practical Tips for Better Prompts

• Write in the present tense. "a cyclist rides through rain" works better than "a cyclist who rides through rain."
• Use specific numbers. "two figures by a campfire" beats "some people by a fire."
• Limit the action to one clear movement. Multiple actions split the model's attention.
• Keep sentences short. Long clauses confuse the model and blur the scene.
• Match the syntax of the tool you use. Most tools accept plain English, but each has a preferred prompt style.
• Describe mood with concrete cues: "soft light," "deep shadows," "warm colors." Abstract words like "nice" or "beautiful" do nothing.

Frequently Asked Questions

Start Writing and Refine From There

Writing good AI video prompts comes down to a simple habit: say exactly what you want in plain words, then keep refining the output. The tools keep changing. The skill does not.

Pick one example from this guide, adapt it to your subject, and run it. Then change one detail and run it again. That feedback loop is how you move from random clips to clips that look intentional.

Save your best prompts, reuse the parts that work, and treat every clip as a step rather than a final product. The more you write, the more you learn what your tool actually sees.