A face swap on video is harder than on a photo: the new face has to turn, blink and talk exactly like the original. Here that's done by MiniMax H3 Recast, which puts the new face into every frame and keeps the expressions and head movement of the source.
You need two files: a 5 to 15 second video with one person in the shot, and a face photo that's close up, well lit and not covered. Leave “Keep the hairstyle from the video” on to change only the face, or switch it off to bring the hairstyle over from the photo too.
Choose 768p or 1080p. The audio of the original video is kept, so the person in the clip still speaks in the same voice.
Deepfakes are easy to misuse, so the rules are strict. Before you start, you confirm that the people in the photo and the video have agreed to the swap and that you won't pass the result off as real footage. Videos and photos of children and teenagers, celebrities and politicians, and explicit scenes are not processed. AI checks this before any credits are charged, and a refusal costs nothing.