aireelsegency@gmail.com 
How to Write Prompts for AI Video Generators in 2026
Short answer: Great AI video prompts read like a shot list: subject + action + camera move + lighting + style, in that order — concrete and physical, never vague. “Slow push-in on a ceramic coffee cup, steam rising, morning window light, shallow depth of field” beats “beautiful cinematic coffee video” every time. Below: the formula, copy-paste templates, and fixes for common failures.
The difference between amateur and professional AI video usually isn’t the tool — it’s the prompt. The same generator produces garbage from “make an ad for my store” and gold from a well-built shot description. Prompting for video is a learnable skill, and this guide teaches the complete method.
The prompt is the director’s chair — specificity is your camera crew.
The Shot-List Formula
Every strong video prompt contains these five elements, in this order:
| Element | What to specify | Example |
|---|---|---|
| 1. Subject | The one main thing in frame | “a ceramic coffee cup” |
| 2. Action | One clear motion, with direction | “steam rising slowly” |
| 3. Camera | One camera move | “slow push-in” |
| 4. Light | Source, direction, quality | “soft morning window light from the left” |
| 5. Style | Look + mood, briefly | “photorealistic, shallow depth of field” |
Combined: “A ceramic coffee cup on a wooden table, steam rising slowly, slow push-in, soft morning window light from the left, photorealistic, shallow depth of field.” One move, one light, one style — stated plainly.
The 7 Rules of Video Prompting
- One action per prompt. “A dog running then jumping then barking” confuses the model. One clear motion, one direction.
- One camera move. Push-in OR pan OR orbit — never two. Mixed camera directions produce smeared, seasick footage.
- Name the light physically. “Lit by a desk lamp on the right, cool blue window light behind” gives the model something real to render.
- Be concrete, not poetic. “Steam rising from a ceramic cup” beats “beautiful cinematic coffee video.” Adjectives don’t render; nouns and verbs do.
- State what should NOT happen. If your tool supports negative prompts: “no text, no extra hands, no fast cuts, no morphing.”
- Image-to-video beats text-to-video for accuracy. When the subject matters (your product, your brand), start from a real image and prompt only the motion.
- Generate small, then commit. Run a cheap low-res test first. Fix the wording of the failing beat — don’t rewrite everything.
AI follows instructions literally — vague prompts get vague videos.
Copy-Paste Prompt Templates
Product ad shot
“[Product] on a [surface], [one camera move: slow orbit/push-in], soft studio lighting from the left, [brand color] background, photorealistic, ultra-detailed. No text, no hands.”
Lifestyle scene
“[Person description] [doing one action] in a [location], golden hour sunlight through windows, slow tracking shot, warm cinematic grade, shallow depth of field. No text overlays.”
Text-to-video establishing shot
“Wide shot of [location] at [time of day], [weather/atmosphere], slow aerial drift forward, [style: cinematic/photorealistic], [mood: calm/energetic]. No people, no text.”
Image-to-video motion (product)
“Slow push-in toward the product, subtle rotation, keep the product shape and label exactly as in the image, soft even lighting. No morphing, no extra objects.”
Fixing the 7 Most Common Failures
| Failure | Cause | Fix |
|---|---|---|
| Morphing faces/hands | Too much motion requested | Slow the move; add “no morphing, no extra fingers” |
| Gibberish text | Model tried to render words | “No text” + add titles in editing instead |
| Product looks wrong | Text-to-video guessing | Switch to image-to-video with your photo |
| Flat, fake look | No real light described | Name a physical light source + direction |
| Chaotic motion | Multiple actions/camera moves | One action, one camera move only |
| Style drift between clips | Different wording per clip | Reuse the same style suffix on every prompt |
| Nothing like the brief | Vague prompt | Rewrite with the 5-element formula |
For the tools behind these techniques, see our AI video ad generator guide and our AI video editing software roundup.
Prompting for Different Tool Types
- Text-to-video models (Runway, Kling, Luma, Pika). Full 5-element formula applies. These models invent everything, so your specificity is the only art direction they get. Keep clips short (5–10 sec).
- Image-to-video. Prompt only the motion: “slow push-in, subtle parallax, keep subject identical.” Describing the subject again can cause errors.
- AI avatar tools (HeyGen, Synthesia). The prompt is the script, not a shot description. Write for the ear: short sentences, conversational tone.
- All-in-one ad makers (CapCut, Canva). Start with the script or product info; let the tool draft, then refine with targeted prompts per scene.
Building Your Prompt Library
Don’t start from zero every time. Keep a simple document with three sections: winning prompts (exact wording of anything that produced great output), style suffixes (your standard endings), and failed prompts with fixes. A 20-prompt library turns every new video from a 30-minute experiment into a 5-minute execution.
Frequently Asked Questions
How long should an AI video prompt be?
One to three sentences covering the five elements (subject, action, camera, light, style). Longer isn’t better — competing details confuse the model.
Should I use text-to-video or image-to-video?
Image-to-video whenever the subject must be accurate (products, brands, real places). Text-to-video for creative scenes where you’re exploring ideas.
Why do my AI videos look fake?
Almost always lighting: flat, sourceless light reads as artificial. Name a physical light source with direction and quality, slow the camera move.
Can I keep the same character across clips?
It’s the hardest problem in AI video. Best practices: reuse the exact same style suffix, keep framing similar, and use image-to-video from a fixed reference image.
Do negative prompts really help?
Yes — “no text, no watermark, no extra limbs, no morphing” prevents the most common artifacts. Add them to every prompt as a standard suffix.
The Bottom Line
Video prompting is shot-list writing: subject + action + camera + light + style, concrete and physical. Master the five elements, keep one move per prompt, and test cheap before committing credits.
Want pro-level AI videos without learning prompting?
NAN Marketing Agency writes, generates, and edits AI video ads for small businesses — you approve the concept, we handle everything else.