Recast a clip with more than four people: the four-photo cap

Sume accepts 1 to 4 reference photos for H3 Max Recast, one per new person. What happens with a group shot, and three ways to work within the cap.

4 min readSume
All posts

The cap, in Sume's code

Sume's Recast builder allows between one and four reference image URLs, one per new person, and refuses anything else with the message that Recast requires 1 to 4 reference image URLs. The Video Router docs describe Recast as swapping the people in a source clip for 1 to 4 photos, and the catalog notes say it replaces the main people. The code comment says fal caps the people swapped in at four, one photo per person.

fal's page, read 2026-10-03, says reference images cost nothing extra and the video is billed at $0.30 per second at 768p or $0.45 at 1080p. The cap is about the model, not the price.

Three ways to handle a larger group

  • Swap only the main people. The catalog describes Recast as replacing the main people, so background extras can stay as they are.
  • Use the optional prompt, up to 2,000 characters, to say who becomes whom and what to keep unchanged.
  • Split the work by shot. If the clip has cuts and different people in different shots, trim it into pieces with a few people each and recast them separately. Each piece must be at least 5 seconds.

What not to do

Do not send five photos hoping the extras are ignored; the request is refused before any charge. Do not send a group photo as one reference; the rule is one photo per person.

Check the output at a few stills before using it. If a person you did not intend to swap changed, narrow the prompt or trim the shot. Sume has no seed for Recast, so another attempt is a new run with a new charge.

Recast people cap, from Sume's catalog and request builder (read 2026-10-03)
ItemValue
Reference photos1 to 4, one per new person
PromptOptional, up to 2,000 characters
Source length5 to 30 seconds
Longest single shot15 seconds

A worked split

Say a 24-second clip has two shots: 12 seconds with three people, then 12 seconds with five. Trim each shot into its own file, recast the first with three photos and the second with the four main people, and join the two outputs. Each piece is above the 5-second minimum, and no single shot is over 15 seconds, so both meet the catalog rules.

The cost is by source second, so splitting does not change the total: two 12-second jobs bill the same seconds as one 24-second job, at fal's $0.30 per second at 768p before Sume's margin.

Sources

Related posts

More in Models

All Models posts

Written by Sume