Videos

Package ai is Hanzo AI — the model API on /v1 (/v1/chat/completions, /v1/messages, /v1/models and the rest of hanzoai/ai's surface) — mounted into a cloud binary with the money, ingest and telemetry callbacks cloud BUILDS but cannot INSTALL.

videos

GET /v1/videos/{id}/content

Implements GET /v1/videos/{id}/content — download the finished MP4.

It authenticates + ownership-checks the caller, then proxies the upstream /content endpoint (bounded by the download concurrency ceiling) and streams the raw video bytes back inline. A successful download also bills the job once (for the client that downloads without first polling to completion) — idempotent with the poll path via job.markCompleted.

GET /v1/videos/{id}

Implements GET /v1/videos/{id} — poll a job's status.

It authenticates the caller, verifies they OWN the job (the caller's billing subject must equal the job's), performs ONE upstream status poll, and — the first time the job is observed completed — settles the reservation with the actual cost and records the billable usage event (exactly once). Returns the OpenAI-shaped video object.

POST /v1/videos/generations

Implements POST /v1/videos/generations — the ASYNC create.

Body: {"model": "...", "prompt": "...", "size"?: "1280x720", "seconds"?: int}

It authenticates the caller, resolves the model to its upstream provider via the shared routing table (zen3-video* / wan2-2-t2v-a14b → the spark-video backend), reserves the per-video budget (the balance gate), creates ONE upstream job, registers it in the in-pod store, and returns the OpenAI-shaped video object with status "queued" IMMEDIATELY. The client then polls GET /v1/videos/{id} and downloads GET /v1/videos/{id}/content. Nothing is billed here — the debit lands on completion.

---

All Hanzo APIs · Interactive reference