AI person swap cost per minute of video: Recast on Sume
One minute of footage through H3 Max Recast costs $22.50 at 768p and $33.75 at 1080p on Sume, more if your pieces round up. Piece-length table inside.

One minute of finished footage costs $22.50 at 768p and $33.75 at 1080p when you swap people with H3 Max Recast on Sume, as long as the pieces are whole seconds. Because Recast accepts 5-30 seconds per job, a minute is always at least two jobs, and pieces with fractional lengths round up and cost slightly more.
Agencies and editors usually quote person swaps per minute of footage, so this is the figure to compare against a reshoot day rate.
Why a minute is never one job
H3 Max Recast takes one source video of 5 to 30 seconds, and no single shot in it can be longer than 15 seconds (fal H3 Max Recast API schema). Sume enforces the same 5-30 second window and asks for the inspected source duration with the request. A 60 second edit therefore has to be cut into pieces first; Sume's video trim cuts a [start, end) range from a clip on media.sume.com into a new MP4.
Cut at shot boundaries where you can: fal's limit of no single shot longer than 15 seconds means a 30 second piece has to contain a cut.
Piece length changes the bill
Sume prices on whole source seconds, rounded up, at $0.375 (768p) or $0.5625 (1080p) a second. Pieces that land on whole seconds add up to exactly the per-minute rate. Pieces with a fraction do not.
| Pieces | Billed seconds | 768p total | 1080p total |
|---|---|---|---|
| 2 x 30 s | 60 s | $22.50 | $33.75 |
| 3 x 20 s | 60 s | $22.50 | $33.75 |
| 4 x 15 s | 60 s | $22.50 | $33.75 |
| 6 x 10 s | 60 s | $22.50 | $33.75 |
| 8 x 7.5 s | 64 s | $24.00 | $36.00 |
| 12 x 5 s | 60 s | $22.50 | $33.75 |
Reading the table
Eight pieces of 7.5 seconds cover 60 seconds of footage but bill 64 seconds, because each piece rounds up to 8. Choose cut points that land on whole seconds when you can, or accept the roughly 7 percent premium.
Twelve 5 second pieces cost the same as two 30 second pieces if each is a whole number of seconds, but they mean twelve submissions, twelve results to join and more chances for a person to look slightly different between pieces.
- fal states that reference images are included at no additional charge, and Sume prices only the seconds, so adding a second or third person (up to four photos) does not change the line.
- Joining the swapped pieces back into one file is a Timeline job: it takes ordered video slots and returns one MP4 (Timeline 1.0).
- A piece shorter than 5 seconds cannot be swapped on its own. Merge it with its neighbour before cutting.
A quote you can reuse
For a quote, state the resolution, the number of pieces and the billed seconds, not just the footage length. 'One minute at 768p, six 10 second pieces, $22.50 for the swaps' is a line a client can check against the rate card, and it leaves trim, caption and join jobs as separate items.
Sources
Related posts
More in Pricing
- AI video cost per deliverable: a rate sheet for finance approval
A cost-per-deliverable sheet built from Sume's published rates: 15-second avatar ad, captioned cut, 6-image set, voiced script, music bed, with a retake factor.
- AI video cost per minute by finished length, 15 s to 10 min
Finished AI video costs about $9.24 per minute at 15 s and $7.86 at 10 min with Wan 3.0, because captions and render round up. Full table, two models.
- AI virtual try-on cost per image: GPT Image 2.5 on Sume
What one try-on still costs on Sume: the documented token rates, why quality auto reserves the max price, and a script that reads the live endpoint pricing.
- AI voice agent cost per minute: speech-to-text, LLM and TTS stack
A dated cost sheet for the October 2026 voice stack: MAI-Transcribe-2-Streaming, Mercury Voice, MAI-Voice-2.1-Flash, and Sume's file STT and TTS, per minute.
Written by Sume