How to write prompts for AI video generation
A plain-English guide to writing better AI video prompts (text-to-video, image-to-video, and reference-to-video) with a simple prompt formula, real examples, and the mistakes to skip.
The AI video prompt formula
Every strong prompt answers five quick questions. Run through them in order and you'll write shootable prompts that AI video generators actually understand.
Subject
Who or what is on screen: a red fox, a vintage car, a barista.
Action
What it does: sprinting, drifting, pouring a latte.
Setting
Where it happens: a snowy forest, a neon alley, a sunlit café.
Camera
How it's shot: slow push-in, drone flyover, handheld follow.
Lighting & mood
The feeling: golden hour, moody neon, soft overcast calm.
Prompts for each type of AI video
Text-to-video prompts
With text-to-video you build the whole scene from words, so detail wins. Instead of "a dog running", try "a golden retriever sprinting along a wet beach at sunrise, camera tracking alongside, warm backlight". Name one clear subject, one main action, and the camera move. Piling on five subjects usually muddies the shot.
Example prompt
“A lone lighthouse on a cliff during a storm, waves crashing below, slow aerial orbit, dramatic overcast light.”
Image-to-video prompts
Image-to-video animates a still you provide, so your prompt should describe motion, not the scene. The picture already sets the look. Your words say what moves and how the camera behaves. Keep it about action: "gently zoom in as she turns her head and smiles, hair moving in the breeze".
Example prompt
“Slowly push in on the character while steam rises from the cup and neon signs flicker behind them.”
Reference-to-video prompts
Reference-to-video takes up to three subject images and stages them together in a new clip. Upload clean, well-lit references, then describe the scene they act out. Great for keeping a character, product, or mascot consistent across videos without re-describing them each time.
Example prompt
“The mascot from image 1 waves next to the product from image 2 on a bright studio backdrop, playful bounce.”
Editing prompts: change one thing at a time
The fastest way to a great clip is rarely the first prompt. It's the edits. After a video is generated, you can refine it with short instructions like “make it night time”, “remove the on-screen text”, or “add falling snow”. Because the edit is stateful, everything you didn't mention stays the same, so you can dial in a shot step by step instead of rolling the dice on a brand-new generation.
Common AI video prompt mistakes
- Stuffing keywords instead of directing a shot: write like a director, not a search box.
- Asking for many actions at once: one clear beat per clip reads far cleaner than five.
- Forgetting the camera and lighting: they decide the mood as much as the subject does.
- Starting over to fix one detail: use editing to change a single thing and keep the rest.
AI video prompt FAQ
How do I write a good prompt for AI video generation?
Direct it like a shot: name the subject, the action, the setting, the camera move, and the lighting or mood. Specific, cinematic direction beats a long list of adjectives.
What's the difference between text-to-video, image-to-video, and reference-to-video?
Text-to-video builds a scene from words alone. Image-to-video animates a still you upload, so you describe motion. Reference-to-video uses subject images to keep characters or products consistent in a new scene.
How do I edit an AI-generated video without regenerating it?
Use stateful editing: after a clip is made, ask for one change at a time ("make it night", "remove the text"). The model keeps everything else intact instead of starting from scratch.
How long can AI-generated video clips be?
Clips run up to about 10 seconds with synchronized audio. For longer stories, generate several clips and stitch them together.