Minimax H3 Recast

It performs a cast substitution rather than generating a video from nothing. The portraits you upload take the roles held by the people in your footage, while movement, framing, edits and audio all carry over from the original file.

The clip has to fall between 5 and 30 seconds, and no individual shot may run past 15 seconds. Split longer takes before you upload, because the finished file comes back the same length as the input.

Up to four people, each with a portrait of their own. The tool maps them onto the leads in screen order by default, so arrange your uploads to match; a written instruction can change that assignment.

One sharp, evenly lit, front-facing portrait per person. Half-body framing carries hairstyle and clothing across more accurately than a tight head crop. Avoid sunglasses, hats, hidden faces, group photos, heavy filters and blurry shots.

Preview in low resolution first to confirm the pairing, then render the final piece at 1080p. Since cost follows clip length, that first pass is the cheap way to catch a mismatch.

Skip it when the footage itself must change — different action, different camera work, a different setting — because all of that comes from your source. Results also suffer with dense crowds, distant or tiny faces, fast motion, and objects covering the face after a hard cut. On long takes, or when someone turns to profile, the likeness can drift back toward the original performer.