Swap 3 people in a 7-second clip: Recast costs $2.625 at 768p

Swapping three people in a 7-second source clip with H3 Max Recast on Sume costs 7 x $0.375 = $2.625 at the default 768p, or 7 x $0.5625 = $3.9375 at 1080p. The price follows the length of the source video, not the number of photos: the Sume estimator multiplies whole seconds by one per-second rate, and one to four person photos ride along for the same rate.

How the rate is built

H3 Max Recast replaces the people in a source video with the people from your reference photos, one photo for each person. Sume bills provider list x 1.25. The provider list is $0.30 per second at 768p and $0.45 per second at 1080p, which gives $0.375 and $0.5625 per second. The model id is h3-max-recast and it accepts source clips of 5 to 30 seconds, so a 7-second clip sits just above the 5-second floor.

Three photos, one clip

Put the source clip in as a public HTTPS video URL and add three public HTTPS image URLs, one per new person. Sume accepts one to four photos. The prompt is optional; if you add one it can be up to 2,000 characters. Recast does not take an aspect ratio, an audio flag, a first frame or reference audio, so the output keeps the framing of the source.

• Source video: 5 to 30 seconds, public HTTPS.
• Photos: 1 to 4, one for each person who should appear.
• Resolution: 768p by default, 1080p if you ask for it.
• Output length equals the source length, so the duration you declare is the inspected source length.

Budget before you submit

Sume reserves the estimated amount when the job is submitted and rounds the reservation up to whole cents, so a $2.625 job reserves $2.63. The catalog states that failed or pre-generation canceled jobs are refunded before capture. If you are testing which of three photos works best, the cheapest loop is to stay at 768p and move to 1080p only for the take you keep: three 768p trials cost 3 x $2.625 = $7.875, while three 1080p trials would cost $11.8125.

Measure the source length first. A 7.4-second clip is a 7.4-second source; check how the estimator rounds before you rely on a figure to the cent, because the pricing function rounds up to whole seconds before multiplying.

When Recast is the wrong tool

If you want to put one avatar face on a clip of 15 seconds or less, the beta Avatar Face Swap route is built for that and is priced per second by quality tier. If you want a photo to copy the movement of a driving video, that is Kling 3.0 Motion Control at $0.1575 per output second. Recast is for swapping the people who are already in the footage.

A worked budget for ten variants

Suppose you want ten different casts of the same 7-second demo. At 768p that is 10 x $2.625 = $26.25, and at 1080p it is 10 x $3.9375 = $39.375. Because the reservation is rounded up to whole cents per job, the ledger holds 10 x $2.63 = $26.30 for the 768p batch and 10 x $3.94 = $39.40 for the 1080p batch while the jobs run. That is a small gap, but it matters if your balance is close to the total.

Keep the three photos consistent between variants: a front-facing, evenly lit photo of one person each. The docs ask for one photo for each person, and give no minimum image size, so I would not guess at one. Check the first result before you queue the rest.

Because the rate depends on the source length alone, shortening the source is the only way to lower the price. A 5-second cut of the same scene costs 5 x $0.375 = $1.875, which is $0.75 less than the 7-second version.