Recast with four people costs the same as one: photos are free

On Sume, replacing four people in a Recast clip costs exactly what replacing one does, because h3-max-recast bills the source length and nothing else. fal lists Recast at $0.30 per output second at 768p and $0.45 at 1080p and says reference images are included at no additional charge (read 2026-10-03). Through Sume's 1.25 multiplier a 10-second source is $3.75 at 768p whether you send one photo or four (Video Router docs, read 2026-10-03).

That is not true of every video model in the catalog, which is why it is worth stating plainly before you budget a multi-person scene.

Why this is unusual

Many reference-conditioned video models add a per-image charge once you pass a threshold. The Video Router docs note, for minimax-h3, that reference-to-video adds $0.08 per image after the fifth image at list. Recast has no such term: the price is a function of seconds and resolution. A two-person dialogue and a four-person group shot cost the same per second, so cost is never a reason to leave a person unswapped.

The arithmetic at four lengths

Because photos are free, the table is a function of length and resolution only. Multiply any row by the number of variants you plan to make, not by the number of people in each.

What does cost you

If photos are free, the real costs are elsewhere. A group scene is harder to track, so you may need more rerenders to get a clean take, and each rerender is a new full-price job. Recast takes no seed, so a rerun is a new draw. Budget two or three attempts for a four-person scene at 768p, and only render the approved take at 1080p.

The photos themselves have a cost of effort: one clear photo per person, in the order the people stand across the frame, because by default the new people replace the main people from left to right. The prompt can override that mapping when the order is not obvious.

• Budget by source seconds times attempts.
• Keep drafts at 768p.
• Order photos left to right or say who becomes whom in the prompt.

A request with four photos

This request replaces four people in a 15-second clip. The bill is the same as for one photo: 15 seconds at 768p, $5.63 through Sume.

curl -X POST https://api.sume.com/v1/video-router/generate \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: recast-four-001" \
  -d '{
    "model": "h3-max-recast",
    "video_url": "https://example.com/group-scene.mp4",
    "reference_image_urls": [
      "https://example.com/p1.jpg", "https://example.com/p2.jpg",
      "https://example.com/p3.jpg", "https://example.com/p4.jpg"
    ],
    "resolution": "768p"
  }'

A group-scene checklist

A four-person swap fails differently from a one-person swap, so check for the failure modes before you spend on 1080p. Look for people who cross each other mid-shot, because the replacement has to follow the right person through the occlusion. Look for people who enter or leave the frame, because the left-to-right default mapping is defined on the main people and an extra can confuse it. And look for faces that are small in the frame; a photo-to-person mapping is easier to verify when each face is large enough to compare.

If a source has those traits, run the 768p draft first and read it closely. At $0.375 per second through Sume, a 10-second draft is $3.75, which is a cheap way to learn that the scene needs a different cut. When the draft shows two people trading places, add a sentence to the prompt naming who becomes whom; the prompt is optional, up to 2,000 characters, and exists for this kind of disambiguation.

A limit you cannot buy past

Four is the ceiling. The catalog accepts 1 to 4 photos, so a crowd scene with eight foreground people cannot be fully swapped in one pass. The practical answer is to choose which four matter, or to cut to shots where fewer people appear. For ordering and prompts see photo order for two to four people.