ElevenLabs Avatars Flows batch hooks vs Sume idempotent submits

To batch avatar videos across products, languages and hooks in Sume, loop over your matrix and submit one async POST /v1/avatar-1.0/talking-video per cell with a unique idempotency key. ElevenLabs advertises a Flows integration for the same kind of batch production inside its tool; Sume leaves the loop in your code or agent, which makes retries and limits explicit.

What ElevenLabs describes

The ElevenLabs page says Avatars integrates with Flows to produce batches across products, languages and hooks. That is a workflow inside the vendor's interface; the Avatars API is stated as planned.

The matrix in Sume

Give each cell its own key, for example product, language and hook joined into one string, so a repeat run does not create duplicates. Each request names the same avatar handle and a script for that cell.

Admission is plan based. Concurrency is Free 1, Pro 4, Startup 8, Scale 20 and Enterprise 20, with queue capacity of max(3, concurrency x 5). Accepted capacity is concurrency plus queue; beyond it you receive 429 queue_full. See generation admission.

Handle the edges

Treat 429 queue_full as a signal to wait, not to resubmit blindly, and treat 402 insufficient_credits as a stop. Use a webhook for terminal events rather than polling every job.

For agents, dry_run=true shows admission and cost first, which is useful before a large matrix goes out.