10 Wan 3.0 clips of 30 s: $18.75 at 480p, $75.00 at 1080p

Ten 30-second Wan 3.0 clips cost $18.75 at 480p, $37.50 at 720p and $75.00 at 1080p on Sume. That is 300 seconds of video; the rate is flat per second for a resolution, so the batch is ten times one clip. If you submit all ten at once, the balance has to cover the sum of the reserves, or the first call that does not fit fails with 402 insufficient_credits.

Batch totals

A draft-then-final mix

Most batches are not ten finals. If you draft all ten at 480p and re-render the three best at 1080p, the bill is the drafts plus the finals.

What must be in the balance

Sume reserves each job's price when you submit it, as Generation admission describes. For a ten-clip 1080p run, up to $75.00 can be on hold while the jobs are queued and running, and the actual charge is settled when each job finishes. Failed generations return their hold.

A plan with a smaller accepted-job queue may reject some submits with 429 queue_full even when the balance is large. Submit in groups and retry rejected calls with their own idempotency keys.

Tracking the batch

Record the job id, the resolution and the idempotency key for each clip in a file as you submit. When the jobs complete, each job's usage.cost can be summed against the plan above, so you can see whether the 480p drafts and 1080p finals matched your estimate.

A request template

Use one key per clip, not one per batch:

curl -X POST https://api.sume.com/v1/videos \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: batch07-clip03-480p" \
  -d '{"model":"wan-3.0","prompt":"A slow dolly across a bakery counter at sunrise","duration":30,"resolution":"480p","aspect_ratio":"9:16"}'

Plan limits for ten at once

Ten 30-second clips are 300 seconds of work. Generation concurrency by plan is Free 1, Pro 4, Startup 8 and Scale 20, with accepted-job capacity of concurrency plus queue (Pro: 4 + 20 = 24). A batch of ten fits the accepted-job capacity on every plan except Free, where the queue holds 5 behind 1 running job, so six submits fit and the rest return 429 queue_full until jobs finish.