Character swap prompt guide for AI video

How a character swap prompt is built

• Open with the task, "Character swap", then name the references: the clip is Video1, the photos are Image1, Image2 in upload order.
• Point at the character in the clip by place and look: "the character on the left in Video1 (black top)".
• Say what comes from the photo: face, hair, skin tone and build, "from the first frame to the last".
• Choose the clothes: the photo’s, or the clip’s with only the person changed.
• List what stays: framing, cuts, timing, every move and lip movement, camera, background, everyone else.
• Close the doors: no split screen, no extra people, no new text, ignore the photo’s background.

Character swap prompt templates to copy

• Character swap. Follow Video1 for the whole clip: same framing, cuts, timing, moves, expressions and lip movement, same camera, no zoom or pan, same background. The character in Video1 becomes the person in Image1: face, hair, skin tone and build from Image1, from the first frame to the last, wearing the clothes from Image1. Ignore the photo’s background. No split screen, no extra people, no new text.

• Character swap. Follow Video1 for the whole clip: same framing, cuts, timing, moves and camera, same background. The character in Video1 becomes the person in Image1: face, hair, skin tone and build from Image1, from the first frame to the last. Keep the clothes from Video1; only the person changes. No split screen, no extra people, no new text.

• Character swap. Follow Video1 for the whole clip: same framing, cuts, timing, moves and camera, same background. The character on the left in Video1 becomes the person in Image1; the character on the right becomes the person in Image2. Each keeps the face, hair, skin tone and build of their photo from the first frame to the last. Ignore the photos’ backgrounds. No split screen, no extra people, no new text.

• Character swap. Follow Video1 for the whole clip: same framing, cuts, timing, moves and camera. Only the character in the middle of Video1 becomes the person in Image1, from the first frame to the last. Every other character in Video1 stays exactly as they are. No split screen, no extra people, no new text.

• Character swap. Edit Video1 and replace only the character: keep its framing, cuts, timing, camera, moves, expressions and lip movement, and its background. The character in Video1 is replaced by the person in Image1, from the first frame to the last. No split screen, no extra people, no new text.

How each model reads a character swap prompt

Numbers follow upload order: the first photo is Image1. Edit-mode models change your clip itself, so the character swap prompt says "edit" or "replace"; the others re-shoot the scene, so the character swap prompt must ask for the same camera and moves.

Words that keep a character swap stable

• "From the first frame to the last": without it, the new face can drift back halfway.
• "No zoom or pan": re-shooting models like to add a camera move of their own.
• "Ignore the photo’s background": or the room from your selfie leaks into the character swap.
• "Every other character stays as they are": or the model swaps every face it sees.
• "No new text": models copy captions they half see and invent new ones.

Why a character swap prompt fails

• It describes the new face ("a blonde woman") instead of the character in the clip. Describe who is in the clip; the photo carries the new look.
• Two swaps in one sentence. Write one sentence per character.
• Many fast cuts. No character swap prompt fixes a character whose face is on screen for half a second.
• A photo with two people. The model cannot tell which one you mean; use one person per photo.