Generate images and videos through MuAPI's unified API

This skill wraps MuAPI's unified asynchronous API for image and video generation so an agent can pick from a live catalog of models rather than integrating each provider separately. It reads the current model catalog (GET /api/v1/models) and a chosen model's schema (GET /api/v1/models/{model}) before building any request, submits exactly one generation POST to the model's documented endpoint, polls the prediction result endpoint with a bounded number of attempts, and downloads the finished output over a fresh, key-free request. It does not guess request fields, bundle an SDK, or hide a billable call from the user.

Use it when the user explicitly asks to generate an image or video with MuAPI, or when an existing media workflow needs a hosted async API it can call over authorized HTTPS. It also fits when a workflow needs to compare or switch between current media models without rebuilding the submit-and-poll logic each time. Before any generation, the skill requires confirming the user is authorized to send the prompt and any reference media to a third-party service, obtaining approval since generation may be billable, and confirming the media type, output location, and whether the user wants a single generation or an explicitly approved batch. Do not use it for text chat, for a provider whose current schema hasn't been fetched and checked, or to auto-retry a generation POST after a timeout - the original request may already have been accepted, so the correct move is to poll the existing request ID first (up to a bounded 120 attempts) rather than create a second paid task.

• Input: MUAPI_API_KEY (required, environment only - never pasted into chat, source, or logs), a chosen model name from the live catalog, and only the request fields confirmed by that model's schema (e.g. prompt, duration, resolution, aspect ratio, or an input URL).
• Output: a request_id from the submission response, then a terminal prediction status (completed/succeeded/success or a failure state), and finally a downloaded, validated HTTPS output file (image/video/audio/3D depending on category).

curl --fail --silent --show-error \
  "https://api.muapi.ai/api/v1/models" \
  | jq -r '.models[] | [.name, .category, .endpoint] | @tsv'

MuAPI's REST API (api.muapi.ai): model catalog, per-model schema, generation submission endpoints (one per model, taken from the catalog), and a prediction-result polling endpoint. Requests use x-api-key header authentication and Content-Type: application/json. The catalog's endpoint field already includes the /api/v1/ prefix, so it is appended to the base URL without adding a second version prefix.

Common failure modes the skill documents explicitly: a validation error means the model's schema changed and the payload needs rebuilding from its current required fields; an immediate timeout after POST means the request ID should be preserved and polled before any resubmission is considered; an HTML, JSON, or empty output file means the HTTPS output URL, status, content type, and file size need checking before the result is treated as real generated media; and a download that would route through a CDN must instead use a fresh, header-free request so the API key is never sent to an output host or redirect destination.

Agent workflows that need image, video, audio, or 3D generation through one hosted API instead of a separate integration per provider, and that can obtain user approval before making a billable call.