Veo 3 prompt generator with sound and dialogue

The structure of a Veo 3 prompt

The builder writes the prompt in a predictable order, so you can read it like a shot list:

Close-up of a small bakery kitchen at dawn. An old baker dusts flour off his hands and looks at the camera. Camera: slow dolly-in. Style: vintage 35mm film photo, golden hour sunlight, cozy mood. Audio: SFX: oven door creaking, birds outside, soft radio music in the background. Dialogue: the character says, "Best part of the day." (no subtitles).
Duration: 8s · Aspect ratio: 16:9
Negative prompt: text overlays, distorted hands

The first sentence is the shot and the setting, the second is the action, then come the camera, the visual style, the audio and the dialogue. The last lines carry duration, ratio and things to avoid. Veo also accepts a negative prompt in some interfaces; where it does not, simply keep the line as part of the text or remove it.

Writing sound that Veo can render

Audio is what makes Veo 3 different from silent video models, and vague sound descriptions waste it. Separate three layers in the sound field:

• Sound effects tied to action: footsteps on gravel, a cup placed on a saucer, a door slam.
• Ambience: rain on a window, distant traffic, a crowded café murmur.
• Music, if any: “soft piano in the background” or “no music” when you want only natural sound.

The builder prefixes your text with Audio: SFX:, a labeling convention many Veo users rely on to keep sound separate from visuals. Keep the list short; three or four sounds are easier for the model to synchronize than ten.

Dialogue: short, quoted and without subtitles

Veo 3 can generate speech with matching lip movement. Put the exact words in the dialogue field; the builder wraps them in quotes and adds “(no subtitles)”, because Veo sometimes burns captions into the frame. Practical rules:

• Keep each line short enough to say comfortably within the clip — roughly one sentence for 8 seconds.
• Describe the speaker in the action field (“an old baker”) so the model knows who talks.
• Add delivery in the action or dialogue field: whispered, shouted, laughing.
• Use one speaker per clip for clean lip sync; split conversations into several shots.

Add “subtitles, text overlays” to the avoid field for extra safety. Do not write dialogue that impersonates real people or puts words in their mouths.

From image to Veo clip

Veo can also animate a still image. If you already have a picture, upload it to image to prompt and choose “Veo 3 (image-to-video)”: you get a prompt focused on motion, camera and an audio line; in local mode it starts with “Animate the provided image”. To design the still first, use the image prompt generator. To compare the same shot for other models, switch the model in the AI video prompt generator. Veo is available through Google products such as the Gemini app, Flow and the Gemini API; features and clip lengths depend on your plan, so check Google's documentation.