/veo-3

name: veo-3
description: >-
  Use to write great prompts for Google Veo (Veo 3.1) to generate or animate video for social media —
  the video-prompt-craft mini-skill, the video counterpart to nano-banana. Run when the user wants a
  Veo / AI video prompt, a short video clip for a post (b-roll, product-in-motion, hook visual,
  spokesperson/UGC clip, ad), to animate a still image into video, or a vertical Reel/TikTok/Short
  clip. Reads brand-profile for brand style. Veo's standout is native synchronized audio in one
  pass, plus image-to-video and native 9:16 vertical. Teaches the prompt anatomy, audio prompting,
  the image-to-video pipeline, and the 8-second constraint + extension/stitching. Honest: iterate
  cheap then finalize, disclose AI video (SynthID watermark), never generate real identifiable
  people or copyrighted IP. This is the
  prompt-craft layer — the API/connection and the generate -> upload to WoopSocial Media -> attach
  flow live in tools/integrations/veo.md; the consuming pack is reels-script's veo-prompt-pack.
metadata:
  version: 1.0.0
license: MIT

Veo (video prompt craft)

Write prompts that get cinematic clips out of **Veo** — Google's video model (**Veo 3.1**; Fast/Lite tiers for cheap iteration). This is the **prompt-craft** layer of a three-layer setup, and the **video counterpart to `nano-banana`**:

• **Connection/API** (model IDs, async generate-and-poll, the *generate → upload to WoopSocial Media →

attach* flow) → `tools/integrations/veo.md`.

• **Prompt craft** (this skill) → how to direct the right clip well.
• **In-skill application** → `reels-script`'s veo-prompt-pack.

> Fast-moving area — re-verify model names/specs quarterly.

Reach for Veo when… (match the job)

Its real strengths: **native synchronized audio** (dialogue + lip-sync, SFX, ambient, music — the differentiator), **cinematic realism/physics**, **image-to-video**, **reference-image consistency**, **first/last-frame** control, and **native 9:16 vertical / 4K**. Reach **elsewhere** for a **pure talking-head explainer** (→ avatar tool like `heygen`), **native-4K/multi-shot/motion-transfer at lower cost** (→ `kling`), **HDR/atmospheric mood shots** (→ `luma`), **long continuous video** (stitch/extend or a length-built tool), or **highly stylized/very-high-volume** output. Don't default it to every job. (Details: `references/when-and-how-to-prompt.md`.)

Step 0 — Read the brand + the job

Load `brand-profile.md` (visual style, palette, tone). Identify the **job** (b-roll / product motion / hook visual / spokesperson / ad / animate-a-still) and the **aspect ratio** (9:16 for social).

Step 1 — Direct the shot (natural language, not quality spam)

Describe a shot like a director: **subject · action · scene · camera (type/movement/angle/lens) · lighting/mood · style · audio · timing**. One clear motion per clip. Set **9:16**. Ground in the brand. See `references/when-and-how-to-prompt.md`.

Step 2 — Use the superpower: audio

Veo generates **native synchronized audio** in one pass — describe it explicitly: **dialogue in quotes** (+ who/tone), **SFX**, **ambient**, **music** mood. Treat generated audio as a **draft/guide track** for branded work (record real VO / license music for final) and **verify** lip-sync. Direct the **camera/motion** for the cinematic feel. See `references/audio-and-camera.md`.

Step 3 — Inputs, length, recipes

• **Image-to-video:** animate a still (e.g. a `nano-banana` frame) — it becomes the first frame; you

describe motion + audio. Use **reference images** / **first-and-last frame** for consistency/control.

• **Length:** one generation ≈ **8 seconds** (4/6/8); **hook in the first second**; for longer use

**scene-extension or stitch** clips. Plan short beats.

• Pick a **recipe** for the job (b-roll, product motion, hook, spokesperson, ad, animate-a-still). See

`references/inputs-length-and-recipes.md`.

Step 4 — Iterate cheaply, then verify, disclose, ship

• **Iterate at low-res / Fast** to lock the prompt; **finalize the keeper at high-res/4K** (4K costs

~40–60% more time/$). Off-prompt generations still consume credits — **prompt skill is the cost lever**; video is **async**.

• **Verify** every clip (artifacts, lip-sync, physics) before publishing.
• **Disclose** AI video per platform/region (EU AI Act; TikTok auto-disclosure); every clip carries a

**SynthID** watermark — don't pass it off as real footage.

• **Ship:** generate per the integration guide → **upload to WoopSocial Media → attach** via

`scheduling-and-queue`. WoopSocial doesn't generate video.

Quality bar — self-check

• Did I **match the tool to the job** (and route talking-heads/long-form elsewhere)?
• Is the prompt a **directed shot** (subject/action/camera/lighting/style), **brand-grounded**, **9:16**,

with an **explicit audio cue**?

• Did I plan for **8-second** clips (hook first) and **iterate cheap → finalize**?
• For stills, did I use **image-to-video** (and reference/first-last frame where useful)?
• Did I handle **SynthID + disclosure**, **audio-as-draft**, **verify the output**, and **refuse real

people / IP**?

• Did I point to **`tools/integrations/veo.md`** for the API + WoopSocial flow (no claim WoopSocial

generates video)?

Edge cases & pushback

• **Talking-head explainer / lots of dialogue** → suggest an avatar tool (`heygen`); don't force Veo.
• **"Make a 40-second video"** → ~8s per gen; scene-extend/stitch; plan short beats.
• **"Generate 10 final 4K-with-audio variations now"** → iterate cheap first; 4K/audio is costly + async.
• **"a person, cinematic, 4k, amazing"** → rewrite into a directed shot (subject/camera/lighting/audio).
• **Real person / copyrighted IP / "post as real footage"** → refuse; SynthID + disclosure; offer an

original alternative.

• **"Generate it in WoopSocial"** → WoopSocial doesn't generate; this prompts Veo, then the clip is

uploaded to Media and attached.

Related

• `tools/integrations/veo.md`