How many H3 Max Recast clips does $50 buy on Sume?

A $50 balance buys 26 five-second or 4 thirty-second H3 Max Recast swaps at 768p on Sume ($0.375 a second); 1080p ($0.5625) buys 17 and 2. Full table inside.

5 min readSume
All posts

A $50 balance buys 26 five-second H3 Max Recast swaps at 768p on Sume, or 13 ten-second ones, or 4 thirty-second ones. At 1080p the same $50 buys 17, 8 and 2. The rates are $0.375 a second at 768p and $0.5625 at 1080p, and the clip length is the length of your source video.

This post turns the per-second rate into a count, because a person-swap budget is usually set as a number of clips, not a number of seconds.

Where the per-second rate comes from

fal lists H3 Max Recast at $0.30 a second at 768p and $0.45 at 1080p, billed on output duration, with reference photos included at no charge (fal H3 Max Recast). Sume's catalog entry for h3-max-recast is the same model at list x 1.25: $0.375 and $0.5625 a second. The output keeps the source length, and Sume rounds the inspected source seconds up before pricing.

Sume's default resolution for this model is 768p. fal's own default is 1080p, so a request that omits resolution on Sume prices at the cheaper tier.

Clips per $50

Each cell is a whole number of clips: the largest count whose total stays at or under $50. The remainder stays in the balance.

H3 Max Recast on Sume, clips per $50 balance by source length (rates read 2026-10-03)
Source length768p per clip768p clips in $501080p per clip1080p clips in $50
5 s$1.8826 (spend $48.75)$2.8117 (spend $47.81)
10 s$3.7513 (spend $48.75)$5.638 (spend $45.00)
15 s$5.638 (spend $45.00)$8.445 (spend $42.19)
30 s$11.254 (spend $45.00)$16.882 (spend $33.75)

What the table does not include

Only the swap is counted. Trimming a long source down to a 5-30 second piece, burning captions, or joining swapped pieces with Timeline are separate Sume jobs with their own prices, so a finished ad costs more than the Recast line alone.

Sume accepts source videos of 5-30 seconds. A 4 second clip is rejected, so the cheapest swap you can buy is the 5 second row.

  • Reference photos are free: a four-person swap costs the same as a one-person swap.
  • A failed job releases its reservation where applicable, so a failure does not consume one of the counted clips.
  • The request has no seed field on Sume, so a second take is a second paid job. Budget a re-roll if the first take misses.

Check the live number before a burst

Sume reserves the estimated amount when a job is accepted, and a balance that cannot cover the reservation fails with 402 insufficient_credits. Before submitting dozens of swaps, preview the cost: the hosted MCP paid tools take dry_run and an optional max_spend_usd (MCP tools and gates). The rate card can change, so treat the table as a date-stamped snapshot and re-derive it from the catalog when you plan a campaign.

Turning clips into a budget

A count of clips is only useful with a plan for failures. Not every swap is usable on the first take, so the number you quote a client should be the table value divided by the takes you expect to need. If you expect two takes per final clip, a $50 balance at 768p delivers 13 final five-second clips, not 26.

Resolution is the other lever. Use 768p for drafts and approvals, since it costs two thirds of 1080p, and re-run only the approved clip at 1080p. That keeps the review loop cheap and spends the higher rate once per clip.

Check the balance before a batch rather than during it. An insufficient balance returns 402 at submit and no job is created, so a half-funded batch fails partway instead of running slower.

Sources

Related posts

More in Pricing

All Pricing posts

Written by Sume