MuAPI Media

Overview

Use MuAPI's unified asynchronous API for image and video generation when a workflow needs model choice without a separate integration for every provider. The catalog and each model's input schema are authoritative; this skill does not guess fields, bundle an SDK, or hide a billable generation request. For a capability overview, see the MuAPI AI video API.

When to Use This Skill

• Use when the user explicitly asks to generate an image or video with MuAPI.
• Use when an existing media workflow needs a hosted asynchronous API and can send an authorized HTTPS request.
• Use when the user needs to compare or switch among current media models without changing the surrounding submit-and-poll workflow.
• Do not use this skill for text chat or for a provider whose current schema has not been fetched and checked.

Preconditions

1. Confirm that the user is authorized to send the prompt and any reference media to a third-party service.
2. Explain that generation may be billable and obtain approval immediately before the generation request.
3. Require MUAPI_API_KEY in the environment. Never ask the user to paste it into chat, source files, command history, or logs.
4. Confirm the requested media type, output location, and whether the user wants a single generation or an explicitly approved batch.

API Contract

Generation and prediction requests use:

x-api-key: $MUAPI_API_KEY
Content-Type: application/json

The catalog returns a model name, category, endpoint, and other metadata. The endpoint value already includes /api/v1/; append it to https://api.muapi.ai without adding a second version prefix. Model input and output fields vary, so read the selected model's schema immediately before preparing a request.

How It Works

1. Discover and validate a model

Use a temporary directory for response files and keep credentials out of every file. The catalog lookup is read-only:

workdir=$(mktemp -d "${TMPDIR:-/tmp}/muapi-media.XXXXXX")

curl --fail --silent --show-error \
  --header "Accept: application/json" \
  "https://api.muapi.ai/api/v1/models" \
  --output "$workdir/models.json"

jq -r '
  .models[]
  | select(((.category // "") | ascii_downcase | test("image|video|audio|3d")))
  | [.name, .category, .endpoint]
  | @tsv
' "$workdir/models.json"

Select a model from the current catalog, then fetch its detailed schema. Do not copy a payload from a different model just because the names look similar:

export MUAPI_MODEL="<model name from the catalog>"

curl --fail --silent --show-error \
  --header "Accept: application/json" \
  "https://api.muapi.ai/api/v1/models/${MUAPI_MODEL}" \
  --output "$workdir/model.json"

jq '.input_schema.schemas.input_data' "$workdir/model.json"

Check required fields, types, enum values, size limits, and the output schema. If the model requires an image or video input, use the exact documented field and an authorized HTTPS input URL; do not invent an upload contract.

2. Prepare one reviewed request

Set the key only in the environment and stop before submission if it is absent:

test -n "${MUAPI_API_KEY:-}" || {
  echo "Set MUAPI_API_KEY before generation." >&2
  exit 1
}

export MUAPI_PROMPT="A small paper boat crossing a calm pond at sunrise, steady camera"

# Add only fields confirmed by model.json. This example uses a prompt field;
# many models also require duration, resolution, aspect ratio, or an input URL.
jq -n \
  --arg prompt "$MUAPI_PROMPT" \
  '{prompt: $prompt}' \
  > "$workdir/request.json"

Review the model, requested parameters, destination, and estimated cost with the user. Do not put MUAPI_API_KEY in request.json.

3. Submit exactly once

Resolve the catalog endpoint and make one POST. Do not automatically retry a generation POST after a timeout: the original task may have been accepted.

model_endpoint=$(jq -er --arg name "$MUAPI_MODEL" '
  .models[]
  | select(.name == $name)
  | .endpoint
  | select(type == "string" and startswith("/api/v1/"))

…