Veo 3 Prompt Engineer

Veo 3 Prompt Engineer turns a one-line idea into a structured prompt that exploits the one thing Veo 3 has over Sora and Runway: native audio. Instead of handing you a visual description and leaving you to score the clip in post, it writes the picture and the soundtrack together — spoken dialogue, ambient beds, and music cues — so the model generates a clip that arrives already scored. Built for creators, ad teams, and short-form editors who write prompts for Google's Veo 3 (via Gemini, Flow, or Vertex AI).

When to use it

• You want a talking character whose lip movement matches generated speech.
• You're producing a 6–8s ad or social cut and need diegetic sound (footsteps, rain, a door slam) timed to the action.
• You keep getting silent or mismatched-audio output and don't know which part of the prompt controls sound.
• You're porting a Sora/Runway prompt to Veo 3 and want to add the audio layer it ignores.

What it does

The skill structures a prompt around the elements Veo 3 actually parses: subject, action, scene, camera (shot size, lens, movement), lighting, and style — then adds the audio blocks. It writes dialogue in quotes (Veo voices quoted lines and attempts lip-sync), ambience as an explicit soundscape line, and music as a cue describing genre, tempo, and mood. It also separates the negative-prompt list and respects the model's ~8 second clip limit and 16:9 / 9:16 aspect framing.

Example

curl -fsSL --create-dirs -o ~/.claude/skills/veo-3-prompt-engineer/SKILL.md https://claudeskil.com/skills/veo-3-prompt-engineer/SKILL.md

> Veo 3 prompt: a barista closing up a rainy café at night, says one line to camera

Visual: Medium close-up, 35mm, slow push-in. A barista in an
apron wipes the counter of a dim café, warm tungsten light,
rain streaking the front window. Cinematic, shallow depth of field.
Dialogue: The barista looks up and says, "We're closed — but stay
dry out there."
Ambience: steady rain on glass, a ticking wall clock, distant traffic hiss.
Music: soft lo-fi jazz piano, slow tempo, melancholic.
Negative prompt: subtitles, text overlay, distorted audio, extra fingers.

Tips & gotchas

• Quote your dialogue. Veo treats quoted text as spoken lines; unquoted "he says hello" often becomes on-screen text instead. The skill quotes for you.
• Keep speech short — roughly one short sentence per ~8s clip. Cramming two lines causes rushed or clipped delivery.
• Don't over-stack audio. Loud music plus dense ambience can bury dialogue; the skill keeps one layer dominant.
• Add subtitles and text overlay to the negative prompt — Veo frequently burns in captions otherwise.

FAQ

Does this call the Veo 3 API for me? No. It produces the prompt text. Paste it into Flow, Gemini, or your Vertex AI request — generation and billing happen there.

Can it do multi-speaker dialogue? Yes, it labels each speaker, but Veo handles two voices in one clip inconsistently. For reliable results, keep one speaker per clip and chain shots.

Will it work for Veo 2? Veo 2 has no native audio, so the dialogue/ambience/music blocks are ignored. Use it for Veo 3 (or Veo 3 Fast).