Keep the performance.
Change who is in it.

You can replace the people in a video by giving Higgsfield the footage plus reference images of the faces you want instead, and a prompt that tells it to match their features while leaving the original movement, expressions, lighting and background untouched.

This is a character replacement prompt. You bring two things: a video you have already shot or generated, and reference images of the people you want in it instead. The job of the prompt is to keep everything about the original clip, the movement, the performance, the lighting and the scene, and change only who is on screen. Below is the prompt verbatim, where to run it, and what each instruction in it is doing.

What you need before you start

Two inputs, both named in the prompt itself: the video you want to alter, uploaded as the base clip, and the reference images of the characters who should replace the ones currently in it. The reference images carry the facial features and physical appearance the model is being asked to match, so they do the heavy lifting. Everything else in the prompt is about protecting what was already in the footage.

Where to run it

Higgsfield. The credit under the next line links straight to it, and that is an affiliate link: it never changes what TALC recommends. Here is how the company describes what it does.

Create images, videos, and voice content with Higgsfield AI from text prompts or references.

— Higgsfield

The prompt

Seamlessly replace the characters in the uploaded video with the provided reference images. Accurately match their facial features and physical appearances while preserving the original video’s movements, expressions, lighting, background, and overall visual consistency. Ensure the replacements blend naturally into every frame, maintaining consistent character identity throughout the video.

What each instruction is doing

Four instructions, each guarding something different. “Seamlessly replace the characters in the uploaded video with the provided reference images” sets the task and names where the new identities come from. “Accurately match their facial features and physical appearances” is the likeness test, and it covers build and proportion rather than the face alone. The clause about preserving movements, expressions, lighting, background and overall visual consistency is the protection clause, and it is the longest one on purpose: everything listed in it is something a generator will happily rewrite if you do not pin it down. “Maintaining consistent character identity throughout the video” addresses drift, the failure where the face is right in the first second and has quietly become someone else by the last.

Sources

1. Higgsfield

Drafted with AI under human editorial review. Third-party sources are cited and linked, never rewritten as our own. · TALC AI