To write your first AI video prompt, describe one short moment in five parts: who is in it, the one thing they do, where it happens, how it looks, and what the camera does. Keep it to a few plain sentences and one main movement.
Then set the clip length in the tool, generate, and change one thing at a time until the clip looks right.
- The formula: subject + one action + setting and light + look + camera (shot size and one move).
- One moment per clip. Google’s Veo guide says a short video should focus on a single moment; three events means three clips.
- Order depends on the tool. Veo puts the camera first, Kling the subject first, Seedance the format first. The cheat sheets have each order.
- Say what you want, not what you do not. Runway: “Describe what you want to see rather than what you don’t want.”
- Wrong clip? Change one thing. Then you know which change fixed it.
A video prompt is a one-line shot list for a film crew that cannot ask you questions. Anything you leave out, the crew decides for you, usually with the most ordinary choice.
What goes in a first AI video prompt?
The plain point: five parts, in plain sentences. This formula is this site’s own beginner version of what the vendors’ guides ask for.
- Subject: who or what, described clearly. “A golden retriever puppy”, not “a dog”.
- Action: the one thing they do. “Chases a red ball.”
- Setting and light: where, when, and where the light comes from. “A sunny back garden, late afternoon.”
- Look: the style and mood. “Natural colours, like a home video.”
- Camera: a shot size and one move. “Low-angle tracking shot.” The words are on camera words for AI video.
Google’s Veo guide lists the same ingredients: subject, action and style as the core, with camera, composition, focus and ambiance as optional extras.1
Length and sound usually live in the tool’s settings. Set the length there, so the action you describe fits the clip. If your tool makes sound, see sound in AI video.
Subject, action, setting, look, camera. Five short parts beat one long list of adjectives.
A worked example, built up step by step
The plain point: start tiny and add one part at a time. Runway’s guide recommends exactly this: begin “with a simple prompt that focuses on the most critical motion components”.2
Step 1, subject and action: A golden retriever puppy chases a red ball. Already a video: something moves.
Step 2, setting and light: … across a sunny back garden in late afternoon, long shadows on the grass. Now the tool is not guessing the place.
Step 3, camera: Low-angle tracking shot at the puppy’s height, keeping pace beside it. Now the tool knows where to stand.
Step 4, look: Natural colours, soft and warm, like a family home video. Now the style is yours, not the tool’s default.
Step 5, one guard line: One continuous shot. Runway’s guide suggests “Continuous, seamless shot” if unwanted cuts appear.2
Generate after each step if you can. In this site’s use, that is how you learn what each part actually does in your tool.
What order does each tool’s guide use?
The plain point: the parts are the same everywhere, but each vendor publishes its own order. Write the five parts, then arrange them the way your tool’s guide asks.
| Tool | Order its guide uses | Cheat sheet |
|---|---|---|
| Veo (Google) | Order: camera first, then subject, action, context, style.3 | Sheet: Veo |
| Runway | Order: camera, subject, action, environment, then extras; start with the essential motion and add one thing per try.2 | Sheet: Runway |
| Kling | Order: subject, how it moves, scene; camera, light and atmosphere optional.4 | Sheet: Kling |
| Seedance (ByteDance) | Order: format (length, shape, shots) first, then subject, action, environment, camera, look, timing, and sound and things to avoid last.5 | Sheet: Seedance |
| Higgsfield | Order: pick the camera preset, name the same move in the prompt, then subject and setting.6 | Sheet: Higgsfield |
| Grok Imagine (xAI) | Order: xAI publishes no formula; its own example prompts are one short line, with shape and quality set as settings, not words.7 | Sheet: Grok Imagine |
Starting from a photo? Then the photo already shows the subject and the look, so the prompt describes only the movement. Google: “Your source image already provides the subject, scene, and style.”8 Runway’s image-to-video order is “The camera [motion description] as the subject [action]. [Additional descriptions]”.2 The full how-to is how to turn a photo into a video.
Same five parts, different order per tool. Open your tool’s cheat sheet once and copy its order.
Three starter prompts to copy
The plain point: each one is a single moment with one move. Change the subject and setting to make it yours. These are this site’s own examples, written in Runway’s camera-first shape.
Want it drafted for you? The free prompt generator takes your subject, a style, light, framing and your tool, and writes the words in that tool’s order, in your browser.
What to change if the clip comes out wrong
The plain point: look at what went wrong, change the one part of the prompt that controls it, and run it again. The table is this site’s own advice, with two vendor lines marked.
| You see | Change this |
|---|---|
| Too much happens | Change: cut it to one action. Google says chaining distinct events in one short prompt “often leads to muddled or incomplete videos.”9 |
| The camera does the wrong thing | Change: one move only, named plainly, with a speed word. For a still camera, say it is entirely still. |
| Every run looks different | Change: add the part you left out. If the place changes every time, you never named the place. |
| The face or clothes change | Change: describe the person the same way every time, or use a reference image. See how to fix flicker and morphing. |
| You asked for “no” something and got it | Change: describe what should be there instead. Runway: “Describe what you want to see rather than what you don’t want.”2 |
| It looks fake or plastic | Change: the light and surface words. See why AI output looks fake. |
If three rewrites change nothing, it is probably not a wording problem. Why your generation came out wrong explains how to read the failure.
One change per run. Two changes and a better clip tells you nothing about which one worked.
Questions people ask
How long should an AI video prompt be?
For a first try, two to four plain sentences. Runway recommends starting simple and warns that over-specification “can paradoxically lead to unexpected or unnatural results” (source above). Add detail one part at a time once the basic clip works.
Should I write the clip length in the prompt?
Most tools have a length setting, so use that. Seedance is the exception among the guides above: its order puts the format, including duration, first in the prompt (source in the table).
Can I use the same prompt in every tool?
The five parts carry over; the order and extras do not. Rearrange the parts to match your tool’s guide, using its cheat sheet, and check whether it reads negatives at all on negative prompts by problem.
Where to go next
- “I need the camera words.” Read camera words for AI video.
- “I want the full structure.” See the anatomy of an AI video prompt and what the vendors actually say.
- “Which tool should I use?” Try AI video platforms sorted by job.
- “My clip flickers or morphs.” Use how to fix AI video flicker and morphing.
- “What do these words mean?” See text-to-video, image-to-video and negative prompt.
- “Take me through it step by step.” Follow make your first AI video.
QUICK CHECK · 3 QUESTIONS
Pick an answer to see if you got it. Nothing is recorded.
1. Which five parts does the page’s beginner formula use?
- Subject, action, setting and light, look, camera
- Title, hashtags, music, length, price
- Style, style, style, camera, length
Show the answer
Answer: Subject, action, setting and light, look, camera. The page’s formula is subject, one action, setting and light, look, and camera.
2. Why does the page tell you to check your tool’s cheat sheet?
- Because only some tools accept camera words
- Because each vendor’s guide puts the parts in a different order
- Because cheat sheets list the prices of each tool
Show the answer
Answer: Because each vendor’s guide puts the parts in a different order. The page says the parts are the same everywhere, but each vendor publishes its own order, which the cheat sheets show.
3. Your clip tries to show three events and comes out muddled. What does the page suggest?
- Add more adjectives to each event
- Write the prompt in capital letters
- Cut it to one action, and make the other events separate clips
Show the answer
Answer: Cut it to one action, and make the other events separate clips. The page says a short clip should be one focused moment, so three events means three clips.
- 1Google AI for Developers, Generate videos with Veo 3.1 in Gemini API, read at source 13 Oct 2026subject, “The object, person, animal, or scenery that you want in your video”; action, “What the subject is doing”; camera positioning and motion, composition, focus and lens effects, and ambiance are each marked “[Optional]”. ↑
- 2Runway Academy, Prompting Guide, undated, read at source 13 Oct 2026“We recommend starting with a simple prompt that focuses on the most critical motion components”; “This over-specification can paradoxically lead to unexpected or unnatural results”. ↑afor unwanted cuts, “First, try increasing the duration to iterate for a seamless shot”, then add “Continuous, seamless shot” to the prompt. ↑b“[Camera] shot of [a subject/object] [action] in [environment]. [Supporting component descriptions]”. ↑c“The camera [motion description] as the subject [action]. [Additional descriptions]”. ↑d“Describe what you want to see rather than what you don’t want.” ↑e
- 3Google Cloud Blog, The ultimate prompting guide for Veo 3.1, 16 Oct 2025, read at source 13 Oct 2026“[Cinematography] + [Subject] + [Action] + [Context] + [Style & Ambiance]”. ↑
- 4Kling AI, Text-to-video prompt guide, 24 Nov 2025, read at source 13 Oct 2026“Prompt = Subject(Subject Description)+ Subject Movement + Scene(Scene Description)”, with “(Camera Language + Lighting + Atmosphere)” marked optional. ↑
- 5Dreamina (ByteDance), Seedance 2.5 prompt guide, 18 Aug 2026, read at source 16 Sep 2026 (as cited on this site’s Seedance cheat sheet): Format + Subject + Action + Environment + Camera + Look + Timing + Audio & Constraints. Re-read at source 13 Oct 2026the template ends “Avoid [unwanted cuts, camera behavior, artifacts, extra objects, text].” ↑
- 6Higgsfield, How to Control Camera Movement, Angles, and Lens in AI Video, re-read at source 12 Oct 2026 (as cited on this site’s Higgsfield cheat sheet): “In Cinema Studio this line runs with the camera movement preset, lighting, and lens already set as parameters”.↑
- 7xAI, Video Generation, read at source 16 Sep 2026 (as cited on this site’s Grok Imagine cheat sheet): example prompt “Make the water crash down and slowly pan out the camera”.↑
- 8Google Cloud, Best practices for Veo on Vertex AI, last updated 9 Oct 2026, read at source 13 Oct 2026, under “Prompt for motion only”: “Your source image already provides the subject, scene, and style.”↑
- 9Google Cloud, Best practices for Veo on Vertex AI, last updated 9 Oct 2026, read at source 13 Oct 2026“For short videos, dedicate each prompt to a single, focused moment.” Chaining “multiple distinct events (A then B then C) in one prompt” “often leads to muddled or incomplete videos.” ↑
Read at source 13 Oct 2026: Google AI for Developers, Generate videos with Veo 3.1 in Gemini API; Google Cloud, Best practices for Veo on Vertex AI (last updated 9 Oct 2026); Google Cloud Blog, The ultimate prompting guide for Veo 3.1; Runway Academy, Prompting Guide; Kling AI, Text-to-video prompt guide; Dreamina (ByteDance), Seedance 2.5 prompt guide.
Reused from this site’s cheat sheets, with the read dates recorded there: the Seedance formula (16 Sep 2026), Higgsfield’s camera guide (12 Oct 2026) and xAI’s video generation docs (16 Sep 2026).
The five-part formula, the worked example, the starter prompts and the fix table are this site’s own, not claims from these sources. The film-crew comparison is this site’s own way of explaining.
Free printable AI cheat sheets, and AI news every week.
Sixteen one-page AI guides and checklists in one free PDF, no sign-up. And if you want to keep up, the newsletter covers what changed in AI each week, in plain English, with the sources.
SUBSCRIBE FREE ↗ GET THE CHEAT SHEET PACK