Video Batch
Submit many text-to-video and image-to-video requests under one model in a single call. Every request in a batch is charged at 50% of its single-request price.
POST
Generate many videos in one batch
Video Batch
Generate several videos with one request. A batch takes 3 to 20 requests that all run on the samemodel_id, queues each one as its own video, and returns one batch_id you can poll for all of them together.
Each request in a batch is billed at half the price the matching single endpoint (Text to Video Ultra or Image to Video Ultra) would charge. In exchange, a request that has not finished 8 minutes after submission is failed and refunded automatically.
Supported
model_id values are wan2.2, ltx-2.3, h3-minimax-t2v, h3-minimax-r2v and h3-minimax-start-end-frame. Requests with an init_image take the image-to-video path; the rest take text-to-video. Both kinds can be mixed in one batch.Request
Make aPOST request to below endpoint and pass the required parameters in the request body.
curl
Body
json
requests— 3 to 20 entries. Each entry accepts the same parameters as the single text-to-video / image-to-video endpoint for the chosen model.defaults— optional. Merged under every entry; a value set on an entry wins over the same key indefaults.track_id— an entry’s owntrack_idis used for its webhook and result; entries without one inherit the top-level value.webhook— called once per request as it finishes, exactly as the single endpoints do, with that request’sidandtrack_id.
Validation
The batch is validated as a whole before anything is queued. If any entry is invalid, the entire batch is refused, nothing is queued and nothing is charged. The error names the first failing entry anderrors lists every one:
json
Billing
Every request in a batch is charged at 50% of its single-request price, at submission. The estimate is computed per entry — so a 10-second request costs twice a 5-second one, then both are halved — and the batch is only accepted if your plan allowance or wallet covers the discounted total:json
refunded: true. A request that finishes later than that is not delivered.
Concurrency
A batch of N counts as N queued requests against your plan’s request queue limit. The whole batch must fit in the slots you have free at submission — a plan with 10 slots and 4 videos already processing can submit a batch of at most 6:json
RateLimit-* response headers reflect the full weight.
Response
The response is keyed like a single video response —status, id, output, message, tip — with the batch fields on top. id and batch_id are the same integer, drawn from the same counter as video ids, so a batch id never collides with a video id. Each entry in results is the same body the matching single endpoint would have returned for that request, plus its index and track_id.
json
batch_id, or any single request on Fetch Video with its own id.Body
application/json
Response
Batch accepted, or refused whole (validation_error, insufficient_balance, rate_limited). Always HTTP 200.
- Option 1
- Option 2
Keyed like a single video response (status, id, output, message, tip) with the batch fields on top

