Balance to fill a Sume queue: 15-second avatar clips by plan

To fill a Sume queue with 15-second standard avatar clips you need $16.56 of spendable balance on the Free plan, $66.24 on Pro, $132.48 on Startup and $331.20 on Scale. A queued job is not free while it waits: Sume reserves its estimated cost when it accepts the submit, so a full queue is also a full set of holds.

The numbers below multiply each plan's accepted-job capacity by the price of one 15-second clip at the Avatar 1.0 per-second rate. Use them to decide how much to top up before a bulk submit, not as a limit Sume enforces.

Why does a queued job need money before it runs?

The generation admission doc says a submit creates a durable job when the request is valid, the balance can be reserved, and the workspace still has accepted-job capacity. The job may start at once or wait as queued until a processing slot opens. The reserve is taken at accept time, successful completion captures it, and a failure or cancel releases it.

That means balance is checked per submit, and the checks add up. If you submit 24 clips on Pro, each estimate is held while the earlier ones run, so the last submit sees the balance minus 23 holds.

How many jobs can each plan hold?

Accepted job capacity is the processing concurrency plus the queue capacity, and the docs give the plan defaults. The dashboard Concurrency tab and generation_limits.concurrency_limit are the source of truth for your workspace, because overrides exist, so treat the table as the default map.

Avatar 1.0 standard quality, without a product image, is $0.184 per second, so a 15-second clip is $2.76. At plus quality, the default when quality is omitted, it is $0.245 per second and $3.675 per clip. The right-hand columns multiply by the accepted capacity.

What if the balance is lower than a full queue?

A submit that cannot reserve its estimate fails with 402 insufficient_credits before provider work starts. Jobs accepted earlier keep running, so the practical effect is that your batch stops at the job where the holds pass the balance. With $100 on Pro at standard quality, 36 clips would reserve $99.36, but the plan accepts only 24, so you hit queue_full first.

The reverse also happens. On Startup, $100 funds 36 standard clips, well below the 48 the queue could hold, so the 402 comes first. Compare your balance with the table before you choose the batch size, and size waves with the generation_limits snapshot that each submit returns.

A small script to size your own plan

The check is multiplication, so it fits in a few lines. Swap in your own per-second rate and clip length.

plans = {"free": 6, "pro": 24, "startup": 48, "scale": 120}
per_second = 0.184   # standard, no product image
clip_seconds = 15
per_job = per_second * clip_seconds
for name, jobs in plans.items():
    print(name, jobs, round(jobs * per_job, 2))

Where do I confirm live numbers?

Read GET /v1/balance before a bulk submit and generation_limits on each response. The API pricing page lists current rates. If your estimate and your balance disagree, the balance wins.