How to Swap Multiple Faces in a Group Photo or Video

Pick the easiest scene you can

Multi-face swaps fail more often because of the scene than because of the tool. Before uploading, check the target against three conditions.

• Faces stay separated. Overlapping heads in a tight huddle are the hardest case; a row of people with clear space between them is the easiest.
• Lighting is consistent across the group. Mixed indoor and window light means each person needs a differently matched source photo.
• Nobody is half out of frame. Faces entering and leaving the shot make detection drop in and out between frames.

If your footage fails all three, cut to a segment that passes or reshoot. No setting rescues a dense, backlit huddle.

Build a source set before you open the editor

One source photo per person, prepared in advance, is the biggest time saver on a group edit.

1. Collect one clear, front-facing photo for each person being replaced.
2. Match the head angle to the target where possible — a profile source suits a profile target.
3. Rename each file to the person’s name.
4. Check each photo for even lighting, and for anything covering the face.
5. Confirm each person has agreed to appear. See AI Face Swap Consent for how to ask.

Mapping mistakes are the most common error in group edits, and nearly all of them come from unlabelled source files.

Map one face at a time

Detected faces are listed in the order the model found them — not by position, and not by who they are. Verify each one individually.

1. Upload the target and let detection finish.
2. Open the first detected face and confirm on the preview which person it is.
3. Assign that person’s named source photo.
4. Repeat for each remaining face. Do not bulk-assign one source to everyone unless that is the intended effect.
5. Generate a 2–4 second test before running the full clip.

On a group of five, run the test with two faces mapped. If the mapping is right on two, it is almost always right on five.

The six problems and how to fix them

The wrong person changed

Re-check the mapping before regenerating. Detection order does not match seating order. Test with fewer replacements to isolate which assignment is wrong.

One face looks unstable on head turns

Shorten the segment, or use a source photo with a closer head angle. A frontal source on a profile target is the usual cause.

The lighting looks painted on

The source photo was lit very differently from the scene. Pick a source with light from a similar direction and colour temperature.

A face drops out partway through

That person turned away, left the frame, or was blocked. Cut the clip to the section where they stay visible, or accept the gap.

Two faces got the same source

A bulk-assign happened, or two source files were swapped. Re-map individually using the named files.

The result is stable but the context is misleading

Do not post it. A group edit that implies something untrue about the people or the event cannot be fixed with better settings — change the concept, or get explicit consent for that specific context.

Know when to stop

If you have regenerated the same difficult clip three times without changing an input, stop. Change the source photo, shorten the segment, or pick a different moment. Re-running identical inputs spends credits and returns the same result.

Credits on group edits

VisoPop charges by clip length and quality, not by the number of people. Swapping five faces in one 10-second clip costs the same as swapping one, so a group edit is not inherently more expensive — only a longer or HD clip is.

Next step

See Multiple Face Swap for how the mapping step works, or open Create Swap to run a short test. Our real example shows what a passing result looks like.