Higgsfield UGC makes 4 generations per run: Sume queue by plan

Higgsfield's UGC guide says one run can create up to 4 generations. Sume does not bundle variants into one run: each avatar video is its own job, and your plan sets how many run at once and how many wait. To make four hooks you submit four jobs, which Pro (concurrency 4) can run together.

Plan limits

From the generation admission docs. Concurrency depends on plan only, and the concurrency_limit field in the response is the source of truth. Queue capacity is max(3, concurrency x 5).

What happens at the edge

Accepted capacity is concurrency plus queue. Beyond it the response is 429 queue_full; a rate burst gives 429 rate_limited; and a balance problem gives 402 insufficient_credits. Free can take 6 accepted jobs (1 running, 5 queued), and Pro 24.

Each variant costs the same as a normal clip, since variants are separate jobs: a 15 second standard video is $2.76 each.

Variant plan

Vary the opening line, keep the avatar and body copy the same, and send an idempotency key per variant. Compare retention after publishing, not by looking at the render alone.

If you need more than the accepted capacity at once, submit in waves and wait for terminal events from a webhook.