How to write a Veo 3.1 prompt — structure & examples

Veo 3.1 is the first mainstream video model that treats audio as a first-class part of the prompt, not an afterthought. If you're still writing prompts the way you did for earlier video models, you're leaving most of Veo's capability on the table.

The core structure Veo 3.1 responds to

Veo 3.1 prompts perform best when they follow a consistent order: subject → action → setting → camera → lighting → sound. Skipping sound entirely is the single most common mistake — Veo will generate silent or generic ambient audio by default if you don't specify it.

[Subject] a woman in a red coat
[Action] walks briskly across a rain-slicked street
[Setting] downtown at night, neon signs reflecting in puddles
[Camera] tracking shot, low angle, slight handheld motion
[Lighting] cool blue neon key light, warm practicals in background
[Sound] rain on pavement, distant traffic hum, her heels clicking in rhythm with her steps

Why sound cues matter this much

Veo 3.1 can generate diegetic sound (sound that exists within the scene — footsteps, rain, dialogue) and ambient sound (background atmosphere) simultaneously. Prompts that specify both produce noticeably more finished-feeling output than prompts that only describe the visual.

Two sound layers to always consider:

• Diegetic: sounds your subject would directly cause or hear — footsteps, a door closing, dialogue, an object breaking
• Ambient: the environment's baseline sound — traffic, wind, room tone, crowd murmur

Camera language that actually works

Veo 3.1 understands standard film-set camera vocabulary far better than vague descriptions. Use terms like:

• tracking shot, dolly in, static locked-off, handheld
• low angle, high angle, eye-level
• shallow depth of field, rack focus

Avoid vague phrasing like "cool camera movement" — it has nothing concrete to act on.

A second example: dialogue scene

[Subject] two friends sitting at a café table
[Action] one leans forward laughing, the other sips coffee and smiles
[Setting] outdoor café, golden hour, string lights overhead
[Camera] static medium shot, eye-level, shallow depth of field
[Lighting] warm golden hour backlight, soft fill on faces
[Sound] café ambience — cups clinking, distant conversation, one line of dialogue: "You're kidding me."

Common mistakes that weaken Veo 3.1 output

1. No sound layer at all — you get generic or silent audio by default
2. Vague camera direction — "nice shot" isn't camera language Veo can act on
3. Overloading a single shot — too many simultaneous actions confuses subject continuity
4. Mixing multiple unrelated ideas in one prompt — split multi-beat scenes into separate shots instead

The takeaway

Veo 3.1's biggest edge over earlier video models is native sound generation — and it's also the part most people forget to prompt for. Structure your prompt as subject → action → setting → camera → lighting → sound, and you'll get noticeably more usable output on the first try.