H3 Max Recast: a 30-second, 4-person swap costs $11.25 at 768p
Sume bills a 30-second H3 Max Recast at 768p for $11.25 and 1080p for $16.88. Reference photos add nothing; fal's page says the same. Arithmetic inside.

A 30-second H3 Max Recast job on Sume costs $11.25 at 768p and $16.88 at 1080p, the same whether you send one reference photo or four. The rate is $0.375 per output second at 768p and $0.5625 at 1080p, which is the fal list price times Sume's 1.25 margin. Add the 5.5% platform fee from Sume's pricing policy and the all-in figures are about $11.87 and $17.80.
H3 Max Recast is the person-swap row in the video catalog: you send a source clip with video_url and one reference photo per new person (1 to 4), and the output keeps the source motion, camera, cuts and sound. The source must be 5 to 30 seconds and no single shot can run longer than 15 seconds.
Sume's constants for this row are dated 2026-10-01 in the repo (fal_h3_max_recast_2026-10-01), the same day fal's list took effect for it, so the table below matches today's page.
The numbers, read 2026-10-08
The table is produced by a script that imports the repo's estimateH3MaxRecastProviderCost and multiplies the fal list by the house margin. Seconds are rounded up before pricing.
| Resolution | fal list per second | Sume per second (x 1.25) | 30 seconds on Sume | With 5.5% fee |
|---|---|---|---|---|
| 768p (default) | $0.30 | $0.375 | $11.25 | $11.87 |
| 1080p | $0.45 | $0.5625 | $16.88 | $17.80 |
The arithmetic
At 768p: $0.30 x 1.25 = $0.375 per second, and $0.375 x 30 = $11.25. At 1080p: $0.45 x 1.25 = $0.5625 per second, and $0.5625 x 30 = $16.88. A 22.3-second source rounds up to 23 seconds first, so it bills 23 x $0.375 = $8.63 at 768p.
fal's model page, read today, lists $0.30 per second at 768p and $0.45 at 1080p and says reference images are included at no additional charge. Sume's estimator has no per-reference line either, so four swapped faces cost the same as one. That differs from MiniMax H3 text-to-video, where reference-to-video adds $0.08 per image after the fifth.
Steps to get a firm number
- Measure the source length in seconds and round up. Anything under 5 seconds is rejected and anything over 30 seconds needs splitting at a cut.
- Check that no shot runs past 15 seconds; if one does, trim the source first.
- Pick 768p for review cuts and 1080p only for the winner. The price gap is 1.5x per second.
- Submit on
POST /v1/video-router/generatewithmodel: "h3-max-recast",video_urland onereference_image_urlsentry per person.
If you move up a resolution
At 1080p the Recast rate is $0.5625 per second, so the same 30-second swap is 30 x $0.5625 = $16.88, which is $5.63 (50%) more than the 768p price of $11.25. Add the 5.5% platform fee and the two totals become $11.87 and $17.80. Use 768p while you check the identity and motion, and spend the extra $5.63 only when the swap is approved.
What Sume does not do
Sume does not offer text-only generation on this row, does not take an aspect ratio, and does not auto-route to it: sume/auto never picks Recast, so you must request it by id. Prices are read from the catalog on 2026-10-08 and can move when fal changes its list.
Sources
Related posts
More in Pricing
- What do 10,000 Haiku 5.5 agent turns cost? 8k vs 120k prompts
Worked token math from Anthropic's Haiku 5.5 price page: 10,000 turns of 8k tokens cost $10.50, while 10,000 turns of 120k tokens cost $625.
- Haiku 5.5 charges more above 100k tokens: keep Sume tools lean
Claude Haiku 5.5 prices prompts over 100,000 tokens at five times the base rate. How a long Sume tool list and job history can push you over.
- Haiku 5.5 Batch API is half price: Sume renders are not batched
Claude's Batch API gives Haiku 5.5 a 50% discount, but a Sume render is billed by Sume. How to batch the planning step and keep caps on the render.
- Higgsfield Genjutsu: 20-second motion transfer is $7.95 at 480p
Sume prices Higgsfield Genjutsu Motion Transfer by the length of your input video: 20 seconds is $7.95 at 480p and $17.03 at 720p. Arithmetic and limits.
Written by Sume