PhotoHero

How to Swap Multiple Faces in a Group Photo

Most face swappers are built around one subject. Give them a group photo and you end up exporting the image once per person and stitching the results together, which is slow and tends to show at the seams.

Swapping everyone in a single pass is a different job, and the hard parts are not the ones people expect. Here is how it works and where group shots actually go wrong.

Steps

  1. 1

    Upload the group photo

    Drop in the photo. Every face in the frame is detected at once, not just the main subject.

  2. 2

    Tap the people you want changed

    Select faces visually. Anyone you do not select is left untouched, pixel for pixel.

  3. 3

    Give each person a face

    Assign a source photo per person, so different people can get different faces in the same run.

  4. 4

    Download one finished image

    Every selected swap is applied in a single pass — no exporting once per person and stitching.

Pick the people, not an ordering

The common failure in multi-face tools is index-based selection: you are asked for "face 1" and "face 2" without being shown which is which, and you only find out the tool disagreed with you after it has generated. In a photo with two similar-looking people that is close to a coin flip.

Choosing targets visually — clicking the face you mean — removes the guess. It also means a mistake is cheap to correct: reassign that one person and run again rather than starting over.

Swapping everyone is usually the wrong call

In a group photo people are rarely all facing the camera the same way. Someone turned three-quarters away will swap less convincingly than the person looking straight down the lens, using the exact same source face. That is a property of how much of the target face is visible, not a quality setting you can raise.

The composite almost always reads better if you are selective: swap the front-facing people, leave the ones turned away or partly hidden. A photo with three convincing swaps and two untouched faces looks real. One with five swaps of uneven quality does not.

Resolution is the limit you will actually hit

A single run handles up to 10 faces, but in practice the count is rarely what stops you — pixels are. Images are normalized to a 1920px long edge, so a face at the back of a wide group shot can be detected and still be resolved into too few pixels to swap convincingly.

If the people who matter are small in frame, crop tighter before uploading. You will get a markedly better result from a tight crop of four people than from the full-width shot of twelve.

Give each person a source shot from a similar angle

Matching head tilt and angle between the source face and the person you are replacing does more for realism than anything else you control. A front-facing source on a front-facing target blends almost invisibly; the same source on a profile does not.

When you are swapping several people at once this compounds, because each pairing is judged independently by anyone looking at the photo. It is worth picking a different source photo of the same person for a differently-angled target.

Ready to try it?

100 free credits on signup. No card required.

Swap multiple faces free

FAQ

Can I put a different face on each person?+

Yes. Each selected face is assigned its own source, so you can map a different person onto every target in the same photo.

How many faces can I swap at once?+

Up to 10 in one run. The practical limit is usually resolution rather than the count — small faces in a wide shot may be detected but lack the detail to swap well.

Can I leave some people alone?+

Yes, and it is the normal case. Only the faces you select are changed; everyone else is untouched.

Do I need an account?+

You can swap multiple faces free with no account, up to a small daily limit, and those results come back watermarked. A free account removes the mark, carries your existing swap across clean, and includes 100 credits.

Why does one person look worse than the others?+

Almost always the angle. A face turned away from the camera exposes less for the AI to work with. Either swap that person with a source shot from a similar angle, or leave them unswapped.

Related