Video & social · The What2Prompt engine

AI video prompt generator for Veo 3, Kling & Seedance

This AI video prompt generator turns an idea into a clear shot brief: subject, action, camera, light and final frame. At the time of writing (November 2026), video tools to consider include Google Veo 3.1, Kling, Seedance, Runway, Wan and Grok Imagine. Adapt each brief to the controls your chosen tool actually offers.

See all prompt guides →
2Which AI will you paste it into?
Short and punchy.
1 credit per prompt

Works with ChatGPT, Claude, Gemini, Midjourney, Suno and more.

Copy · Paste · Done
STEP 3 · YOUR PROMPT
Prompt for Video AIQuickExample

Role

Video shot brief.

Context

One short shot or a sequence of separately generated shots.

Task

A single continuous shot of a miniature paper city unfolding on a wooden desk.
Begin close to a folded paper square; buildings rise one after another until a small skyline fills the center.
The camera slowly pulls back without cutting.
Warm desk-lamp light, realistic paper texture, quiet home-studio background, deliberate gentle movement and a clear final frame.
Aim for an eight-second scene if that duration is supported.
Keep the action focused on unfolding rather than adding characters or another location.

Format

Separate subject motion, camera direction, lighting and final frame.

Guardrails

Use only supported facts and supplied material. Flag uncertainty rather than inventing evidence or promising results.
5 free prompts

Example prompts

Example

One continuous reveal

Role

Video shot brief.

Context

One short shot or a sequence of separately generated shots.

Task

A single continuous shot of a miniature paper city unfolding on a wooden desk.
Begin close to a folded paper square; buildings rise one after another until a small skyline fills the center.
The camera slowly pulls back without cutting.
Warm desk-lamp light, realistic paper texture, quiet home-studio background, deliberate gentle movement and a clear final frame.
Aim for an eight-second scene if that duration is supported.
Keep the action focused on unfolding rather than adding characters or another location.

Format

Separate subject motion, camera direction, lighting and final frame.

Guardrails

Use only supported facts and supplied material. Flag uncertainty rather than inventing evidence or promising results.
Example

A clean product-style motion shot

Role

Video shot brief.

Context

One short shot or a sequence of separately generated shots.

Task

One red ceramic mug sits on a simple kitchen counter in morning window light.
A hand enters from the right and places a teaspoon beside it; steam rises steadily from the mug.
Locked medium close-up, shallow depth of field, realistic proportions and restrained movement.
The shot should feel calm and usable as background footage for a coffee tutorial.
Keep any sound direction separate and only apply it if the selected tool supports audio.

Format

Separate subject motion, camera direction, lighting and final frame.

Guardrails

No scene transition or extra hands.
Example

Maintain continuity across shots

Role

Video shot brief.

Context

One short shot or a sequence of separately generated shots.

Task

Plan three separate short shots of the same imaginary floating garden for a travel-style montage.
Establish the garden’s appearance first: moss-covered stone platforms, pink flowers, narrow wooden bridges and soft afternoon light.
Repeat essential visual identifiers consistently.

Format

Then write one prompt each for a wide establishing view, a slow bridge-level move and a close-up of flowers in the breeze.

Guardrails

Explain where reference images could help continuity, without guaranteeing that text alone preserves every detail.

Google Veo 3.1: separate scene, camera and audio

At the time of writing (November 2026), Google’s Veo 3.1 guidance describes subject, action, setting, camera and audio as useful prompt elements. Start with one scene: a ceramic mug on a counter, steam rising, a slow push-in and soft window light. Put sound direction in a separate sentence if your selected mode supports it. Check the available duration and reference controls rather than assuming the prompt overrides interface settings.

Kling: make the interaction observable

For a Kling brief, describe the subject’s motion separately from the camera. “A hand places a spoon beside the mug; camera remains locked” is easier to inspect than “cinematic coffee magic.” State the starting position, one interaction and the final pose. If your chosen mode accepts references, identify what each one should preserve. Review hand anatomy, object contact and continuity; adding more adjectives does not guarantee correct motion.

Seedance: give reference assets explicit jobs

When using reference inputs in Seedance, explain which image supplies the subject, which supplies the setting and which movement you want. Do not assume every uploaded asset controls all aspects of the clip. A short sequence needs a clear order and a consistent subject description. Specify the actual supported workflow before requesting a complex timeline, and split an overloaded idea into individual shots you can review.

Runway: describe visible motion instead of abstract mood

A Runway shot brief should make the change visible: fabric lifts in a breeze, a character turns toward the doorway or a camera tracks beside a bicycle. Give one primary movement and a clear composition. For image-to-video, focus on what moves in the supplied frame instead of redescribing everything already visible. Use the official guidance for your selected mode; controls and prompt behavior can differ between releases.

Image-to-video prompts: preserve the starting frame

Use a reference image you own or have permission to process. Identify the subject that must remain recognizable and request a restrained movement: steam rises, the camera advances slightly and the mug stays unchanged. Avoid asking the same short clip to rotate the product, change its material and move to a new location. Inspect the opening, middle and final frame for identity drift before using the clip.

For Wan or Grok Imagine, the same subject/action/camera framework is a starting point, not a promise of identical features. Check the selected product’s current inputs and limits. What2Prompt prepares AI video prompts; it does not render footage or automatically send references to the named services.

What makes a good AI video prompt

  1. Keep one dominant action per shot. Several transformations and location changes compete for a short runtime and make revisions difficult.

  2. Specify framing and camera movement separately from subject motion. A moving subject does not imply the camera should also orbit or zoom.

  3. Name the important visual identifiers for continuity. Use an available reference workflow when an exact character or product must stay consistent.

  4. Match duration, aspect ratio and audio directions to your actual tool. Veo, Kling and Seedance do not expose identical controls or guarantees.

  5. Describe the opening and ending clearly. A useful final frame makes it easier to cut the clip into a larger video.

Bad prompt vs What2Prompt prompt

Bad prompt

“make an epic cinematic AI video”

What2Prompt prompt

Role

Video shot brief.

Context

One short shot or a sequence of separately generated shots.

Task

Single shot: a folded paper square becomes a miniature city on a desk while the camera slowly pulls back.
Warm lamp light, realistic paper texture, no location change, clear skyline at the end.

Format

Separate subject motion, camera direction, lighting and final frame.

Guardrails

Use only supported facts and supplied material. Flag uncertainty rather than inventing evidence or promising results.

The revised brief trades vague intensity for a visible sequence you can inspect. The What2Prompt engine writes a prompt for your selected video AI; it does not render footage here or promise that every generated motion will match.

Write your first shot as a storyboard sentence before adding style. Check whether the action can plausibly fit the selected duration, then describe lighting and camera choices that support it. After generation, review the opening, the transition and the final frame separately. Google’s prompt design documentation links to dedicated video guidance; check your video tool’s current controls rather than copying model-specific syntax from an unrelated generator.

Reference: Google prompt design strategies ↗

Questions about AI video prompt generator

Can I use the same prompt in Veo, Kling and Seedance?

You can reuse the underlying scene description, but adapt controls and supported features in each tool. Keep subject, action, camera and lighting portable; avoid assuming a duration setting, audio feature or reference-image mode exists everywhere. Compare outputs by the shot’s intended purpose rather than expecting identical results from different models.

Why does an AI clip have strange hands or object movement?

Generated motion can contain visual errors even with a clear prompt. Reduce competing actions, simplify the shot and inspect each output before using it. A locked camera and a single interaction can make defects easier to spot. Do not treat a longer prompt as a guarantee of accurate physics or anatomy.

How do I keep the same character in multiple clips?

Write consistent identifying details and use reference-image or character tools where supported. Plan shots around the same established design instead of rewriting it each time. Text alone may not preserve an exact face, costume or object; review continuity across outputs and revise or replace clips that do not match.

What2Prompt prompt engine, with the W2P mark on cobalt.

Give your next idea a better brief.