Skip to content

106 social media skills for AI agents - strategy, writing, video, design, platform growth, publishing, and analytics. Works with Claude, Cursor, OpenClaw, Hermes & 40+ agents.

4.6 KB
Raw Blame

Veo (Google video) — integration guide

The canonical reference for any skill that generates video with Veo — Google DeepMind's text/image-to-video model. Shared infrastructure: video skills point here instead of restating API details. (reels-script bundles a Veo prompt pack; this guide is the connection layer behind it.)

Model versions, regions, and pricing change — re-verify quarterly against the official docs, which are the source of truth: https://ai.google.dev/gemini-api/docs/video and the Gemini Enterprise Agent Platform (Vertex) Veo docs.

What it is

Veo generates high-fidelity short video clips with natively generated audio (dialogue, SFX, ambience) in a single call. As of June 2026 (verify):

  • Veo 3.1 — veo-3.1-generate-preview; Veo 3.1 Fast — veo-3.1-fast-generate-preview. Current flagship: 8-second clips at 720p/1080p/4K, native audio, landscape 16:9 and portrait 9:16, video/scene extension (chain Veo clips up to ~148 seconds total), first/last-frame control, up to 3 reference images, and improved character consistency across scenes. Still paid preview on the Gemini API; GA endpoints are available via the Gemini Enterprise Agent Platform (see Access).
  • Veo 3.1 Lite — veo-3.1-lite-generate-preview. High-efficiency, low-cost lane for high-volume/at-scale generation. No 4K and no Extension — use the full 3.1 model when you need either.
  • Veo 2 — veo-2.0-generate-001. Stable GA, silent (no native audio); use only if you need the older stable lane.

Access

  • Gemini API — https://generativelanguage.googleapis.com/v1beta, @google/genai SDK (generate_videos). Generation is asynchronous: kick off an operation, poll until done, then download. (Paid preview — a Gemini API key with billing is required.) The Interactions API is now GA and is Google's recommended surface for the latest Veo features/models.
  • Google AI Studio — testing surface (Veo Studio demo app; limited free Veo usage, regional).
  • Gemini Enterprise Agent Platform (formerly Vertex AI) — enterprise/scaled, and where the GA Veo 3.1 endpoints live; regions include us-central1, us-east4, europe-west4, asia-northeast1 (verify). Consumption-based (per-second) billing. Migrate any old preview endpoints to the GA endpoints to avoid interruptions.
  • Flow / Gemini app / YouTube Shorts / Google Vids — no-code surfaces (Veo 3.1 runs in the Gemini app and Flow).

API key from Google AI Studio or Google Cloud Console. Output is MP4 (H.264 / AAC).

Capabilities & limits relevant to social

  • 8s per generation — the hard limit. For longer, extend/stitch clips (each new clip continues from the prior one's final second for continuity) up to ~148s total via Extension.
  • Extension is Veo-only + time-boxed — you can only extend a video Veo generated; generated videos are stored ~2 days (referencing one for extension resets its 2-day timer). Not on Lite.
  • 9:16 vertical — native support for Reels / TikTok / Shorts.
  • Native audio — dialogue and sound without a separate audio pass.
  • Reference images + character consistency — keep a subject stable across shots.
  • Generation time ~60-180s per clip (verify); build retries into agentic flows; failed generations generally aren't charged.

How skills use it

reels-script writes the script as a production doc (shot · on-screen text · spoken · timing) and bundles a Veo prompt pack mapping beats → 8-second prompts (see reels-script/references/veo-prompt-pack.md). The user generates per beat, then assembles/captions (CapCut/editor). Hand the finished video to WoopSocial's Media domain (raw-bytes upload) to attach + schedule (see integrations/woopsocial.md).

Compliance — AI-disclosure

All Veo output carries a SynthID watermark — it is AI-generated video. WoopSocial auto-discloses on TikTok; for other platforms, follow their AI-content/synthetic-media rules and the brand's policy. Note EU AI Act synthetic-media disclosure obligations where relevant. Don't put fabricated claims or fake testimonials in generated video.

Relationship to other pieces

  • This guide = the connection/API layer (shared).
  • veo-3 mini-skill (Tier 2) = prompt craft for cinematic, audio-aware clips.
  • A skill's bundled pack (e.g., reels-script) = how that skill applies it.

Skills that use it

veo-3 (prompt craft, under the ai-video router), reels-script (veo-prompt-pack), tiktok-script, youtube-shorts, scripting-and-storyboarding.