Veo 3 Prompt Generator

What is Veo 3?

Veo 3 is a video generation model from Google DeepMind. Like other text-to-video models it turns a written description into a short clip, and its headline feature is native audio: it can generate dialogue, ambient sound and sound effects together with the picture. Google offers Veo through products such as the Gemini app and Flow, and for developers through the Gemini API and Vertex AI. Newer point releases appear from time to time, and access, clip length and pricing vary by product and plan, so always check Google’s current documentation.

Because Veo 3 handles sound as well as images, a good prompt has more to describe than a silent video prompt. You need to tell the model what we see and what we hear, and keep both in sync. That is where a structured prompt helps.

Why JSON prompts are popular for Veo 3

Veo 3 accepts ordinary text, and Google’s own guidance is written in plain language. JSON is not required. Many creators still like JSON prompts because they force you to fill in each part of the shot separately. It becomes obvious when you forgot the lighting, or when the camera field says “static” while the action says “we follow her down the street”.

JSON is also handy for iteration. You can keep the scene and subject fixed and change only “camera” or “audio” between generations, which makes it easier to understand what each change does. The model reads the content of the fields; the curly braces themselves do not unlock anything special, so treat JSON as an organizing tool rather than a trick.

Recommended Veo 3 prompt fields

scene / setting

Location, time of day, weather and era in one sentence.

subject

Who or what is in frame, with consistent visual details.

action

One clear action that fits a short clip.

camera

Shot size, angle, lens feel and one camera movement.

lighting & style

Light quality and direction, color palette, visual style.

audio

Quoted dialogue with speaker and tone, ambience, and sound effects.

Veo 3 prompt examples

Here is the same structure in JSON and in plain text. Both work; use whichever is easier for you to edit.

{
  "scene": "Sunlit farmhouse kitchen, late morning, flour dust floating in the light",
  "subject": "A grandmother in her seventies with a blue apron and her eight-year-old grandson",
  "action": "She guides his hands as they knead dough together on a wooden table",
  "camera": "Medium close-up at table height, slow push-in, shallow depth of field",
  "lighting": "Warm window light from the left, soft shadows",
  "style": "Photorealistic, gentle documentary feel",
  "audio": {
    "dialogue": "Grandmother, softly: \"Gently, like you are tucking it in.\"",
    "ambience": "Quiet kitchen, birdsong outside the window",
    "sfx": "Dough pressing on wood, a soft laugh from the boy"
  },
  "aspect_ratio": "16:9"
}

A small cyberpunk delivery drone with a glowing orange package weaves between skyscrapers in heavy rain at night. Tracking shot from behind the drone, slightly above it, fast but smooth. Neon signs in pink and teal reflect on wet glass towers; rain streaks catch the light. Cinematic, high contrast, photorealistic. Audio: whirring rotors, heavy rain, distant sirens and a low synth pulse.

Small, dim comedy club with a brick wall and a single spotlight. A comedian in a denim jacket leans into the microphone and delivers a punchline, then pauses with a straight face. Medium shot from the audience’s point of view, static camera, heads of the crowd softly out of focus in the foreground. Warm spotlight, dark surroundings. Audio: the comedian says, "And that is why I don’t lend my car to my dentist." The crowd erupts in laughter and applause.

Tips for better Veo 3 results

• check_circleKeep it to one shot. Clips are short, so describe one moment instead of a full scene with cuts.
• check_circleWrite dialogue the way it should sound: short sentences, natural rhythm, and a delivery note such as “softly” or “excited”.
• check_circleDescribe sound in layers: background ambience, then specific effects that happen on screen.
• check_circleIf you do not want captions or text overlays, say so; video models sometimes add text when dialogue is present.
• check_circleReuse character descriptions word-for-word across prompts to improve consistency between clips.
• check_circleIterate one field at a time and keep notes on which change improved the clip.

Common mistakes

• check_circleConflicting fields, such as “static camera” plus an action that needs the camera to follow the subject.
• check_circleOverloaded audio: five sound effects and two speakers in a short clip often sound muddy.
• check_circleRequesting real public figures, brands or copyrighted characters, which Google’s policies restrict.
• check_circleVery long monologues. A line or two of speech fits short clips much better.
• check_circleAssuming JSON is mandatory. If your tool or workflow prefers text, switch the toggle and the generator writes a natural paragraph instead.

More ways to build prompts

Need full control over every key? The JSON Prompt Generator lets you build custom structures. Want to recreate a clip you already have? Upload it to Video to Prompt and adapt the result for Veo 3. For multi-scene ideas, the Video Prompt Generator writes scene-by-scene breakdowns you can split into separate Veo 3 shots.

Frequently asked questions