Veo 3 Prompting Guide: Audio Direction, Dialogue, and Cinematic Control

What makes Veo 3 different

Every other video model treats audio as a post-production problem. You generate the video, then layer audio on top in an editing tool. Veo 3 breaks from that entirely. Audio is part of the generation itself, built from the same prompt that creates the visuals.

That changes the prompting model completely. You're not describing a clip anymore. You're writing a complete scene brief — visuals, motion, sound design, dialogue, and tone all in one pass.

The other key capability: Veo 3 has exceptional physical realism. Fabric movement, liquid dynamics, particle behavior, and environmental physics render at a level that makes most outputs look shot on a real camera rather than generated. This is the model you use when physical plausibility matters.

The six-layer prompt structure

Veo 3 responds to structured, layered prompts. Each layer handles a different dimension of the output. Use this order every time, and never skip a layer — especially audio.

1. Visual style

Open with the type of video you're creating. This is the container everything else fits into. Veo 3 understands a wide range of styles and will commit to them if you name them early:

• Cinematic realism — looks like a film or prestige TV production
• Documentary footage — handheld, naturalistic, slightly imperfect
• Commercial product film — clean, controlled, brand-ready
• Animated — specify the sub-style: 2D hand-drawn, 3D rendered, claymation, stop-motion
• Archival/texture — VHS tape, Super 8mm, worn 35mm, Polaroid grain

2. Location and environment

Be specific and sensory. Vague locations produce averaged results. Specific locations produce atmosphere:

Weak: "a city street at night"
Strong: "a rain-slicked Tokyo alleyway at 2am, neon signs in kanji casting fractured reflections across wet pavement, steam rising from a grate, no other people in frame"

Include time of day, weather conditions, architectural detail, and distance scale. Veo 3 uses all of it.

3. Characters

Describe each character with full visual specificity. Age, build, clothing, hair, distinguishing features. Use named or labeled identifiers — you'll need them in the audio section to assign dialogue:

[DETECTIVE]: a tired man in his mid-50s, rumpled navy trench coat, dark circles, unshaven, holding a cigarette that's mostly ash

The more visual detail upfront, the more consistent the character stays throughout the clip.

4. Action and motion

Describe what physically happens from start to finish. Always include a motion endpoint. Open-ended actions ("walks around," "looks at things") cause Veo 3 to loop or distort. Close every action:

Open-ended (bad): "she walks through the market"
Closed (good): "she walks through the market, slows at a fruit stall, then stops, picks up an orange, and turns to look directly at the camera"

Temporal language is meaningful to Veo 3. Use "then," "as," "until," "while," and "before" to sequence actions with precision.

5. Camera direction

Without a specified camera move, Veo 3 defaults to static or inconsistent movement. Always specify:

• Dolly in/out — physical push toward or pull away from subject
• Pan left/right — horizontal rotation from fixed position
• Tilt up/down — vertical rotation from fixed position
• Tracking shot — camera follows subject laterally
• Crane up/down — vertical elevation or descent
• Orbit (360 arc) — circles the subject
• Rack focus — shifts focus plane between foreground and background
• Handheld drift — naturalistic slight instability
• Locked off static — completely still, no movement
• POV — first-person perspective from a character or object

Pair camera moves with shot framing: tight close-up, slow dolly in or wide establishing shot, locked off static.

6. Audio direction

This is Veo 3's differentiating layer, and the one most people skip entirely. Every prompt should have explicit audio direction. Break it into three components:

• Dialogue: assign lines to specific characters using the labels from section 3
• Sound effects: name specific sounds tied to actions in the scene
• Ambient and music: describe the environmental soundscape and any music cues

Audio direction in depth

Because audio is Veo 3's biggest capability gap over other models, it deserves its own section.

Formatting dialogue

Use character labels consistently between the character description and the dialogue section. Mismatched labels break lip sync and voice assignment:

You can also specify voice quality directly: gravelly and tired, bright and young, nervous energy, flat and controlled, menacing undertone. Veo 3 maps these voice descriptors onto the generated audio.

Formatting sound effects

Tie sound effects to specific actions in the scene. Name them precisely rather than generally:

Vague: "city sounds"
Specific: "the hiss of a bus door opening two blocks away, distant traffic, one set of footsteps on wet pavement, the creak of a fire escape in wind"

Formatting music cues

Describe music by genre, tempo, instrumentation, and emotional tone. Not just "sad music" — that's the vague version that produces averaged output:

Vague: "suspenseful music"
Specific: "sparse jazz piano, slow tempo, single sustained note underneath, no percussion, minor key, melancholic not dramatic"

Complete prompt example

Here's a full Veo 3 prompt using all six layers:

Style vocabulary that works

What Veo 3 is best at

• Dialogue scenes — lip sync, facial performance, and voice delivery are exceptional compared to any other video model
• Physical realism — liquid, fabric, smoke, particle behavior render with convincing physics
• Atmospheric environments — weather, lighting conditions, and time-of-day are rendered with strong fidelity
• Product shots — commercial-style close-ups with clean light control
• Short narrative scenes — character-driven, single-location dramatic moments

Common failures and how to fix them

• Muted or generic audio: Skipping the AUDIO section or writing vague audio direction ("some background noise") causes Veo 3 to generate low-effort ambient sound. Specific audio direction is not optional.
• Open-ended action loops: Describe where every action ends. "Walks to the window and stops" closes the loop. "Walks around" does not.
• Character label mismatch: If you label a character [VERA] in the description and then reference her as [WOMAN] in the audio section, dialogue sync breaks. Use identical labels throughout.
• Missing camera spec: Omitting camera direction produces static or erratic movement. Name the move explicitly every time.
• Overloaded scene elements: More than 6-8 meaningful elements in a single prompt causes the model to average them into visual mush. Fewer, more specific elements produce sharper results.
• Skipping the style opener: Without a style declaration, Veo 3 defaults to a generic "realistic" look. Name the aesthetic first, every time.

Vertical video for Shorts and Reels

Veo 3 supports 9:16 output for vertical formats. The prompting approach is identical, but composition matters more in vertical. Specify framing with vertical in mind:

• Use vertical 9:16 framing explicitly in the camera direction
• Center the subject in the upper third rather than dead-center for natural vertical composition
• Keep action close and personal: vertical rewards tight shots over wide establishing shots
• Dialogue scenes work exceptionally well in vertical — the format isolates the speaker naturally