Introduction

An agent-facing media generation API — video, images, and captions over one REST surface.

AgentHook is a hosted media-generation API built for AI agents. Claude Code, Cursor, n8n — anything that can run a CLI command or make an HTTP call — can create social-media-ready content: UGC-style talking-head video, images (including reference-driven edits), and burned-in captions.

There is no timeline editor and no render farm to babysit. Your agent describes what it wants, the API returns a durable URL to the finished media.

The tools

Tool	What it does
make_video	Prompt (+ optional reference images) → video with native audio. A person described in the prompt speaks the lines you write. Optionally chains captioning onto the result.
make_image	Prompt (+ optional reference images) → images. With references it edits/composes from them; without, it generates from text.
caption_video	Existing video URL → the same video with styled, burned-in subtitles (movie or tiktok preset).
create_influencer	Rough idea + name → a reusable, account-bound character (portrait + character sheet). Reference it by slug via --influencer on every later make_video and make_image to keep the same person across generations.

How every run works

All three tools share one uniform, async shape:

1. POST /api/v1/tools/{tool}/run with a JSON body → 202 with a run_id. Credits are debited atomically at this moment.
2. Poll GET /api/v1/runs/{run_id} until status is completed or failed.
3. completed → output holds durable URLs to your media. Outputs are stored on our CDN and do not expire.
4. failed → your credits are refunded instantly and automatically. See Credits & refunds.

Run states: queued → processing → completed | failed.

Built for agents

• Self-describing. GET /api/v1/tools serves the full input schema for every tool — names, types, defaults, enums. Wrappers (the CLI today; MCP and a skill pack later) read schemas from the API instead of duplicating them.
• Deterministic validation. The CLI pre-validates inputs locally against those schemas before spending a network call or a credit; the server enforces the same rules authoritatively.
• Searchable history. GET /api/v1/generations?q=... full-text-searches your past runs by prompt, enhanced prompt, and transcript — so an agent can find “that video about the standing desk” without storing anything itself.
• Credit-metered. Prepaid credits, atomic debits, instant refunds on failure, and a full ledger at GET /api/v1/credits/history.

Next steps

Quickstart

Sign up, grab a key, and generate your first talking-head video in minutes.

Authentication

Bearer API keys — how to create and use them.

Tool reference

Every argument, type, default, and enum for the three tools.

Using from Claude Code & Cursor

Hand the CLI to your coding agent.