# How to choose a video that works for an AI person swap

> What makes a clip easy or hard for an AI person swap, from face size and camera motion to crowds and length, with a checklist to use before you upload.

Source: https://vswp.app/guides/choose-a-video-for-ai-swap/ · Updated 2026-10-03

A video works best for an AI person swap when the people you want to replace are large in the frame, clearly lit, and mostly facing the camera, the camera moves calmly, and the best moment fits into 30 seconds. Crowds, faces hidden behind hands or props, and very fast motion make a good result harder. Pick the clip with the same care as the photo, and you'll need fewer takes.

## Why does the choice of video matter so much?

The AI keeps the original video's moves, camera, lighting and sound and redraws the people you replace in every frame. The more clearly a person is visible, the more the AI has to work with. When a face is tiny, turned away, or half covered, it has to guess, and guesses are where the likeness slips.

That's why two people with the same photo can get very different results from two different clips. The photo matters, as our guide on [which photos work best for an AI video swap](/guides/best-photos-for-ai-video-swap/) explains, but the clip sets the ceiling.

## What does a good clip look like?

Check these before you upload:

- **Large faces.** The people you replace should fill a good part of the frame for most of the clip, not appear as small figures in the distance.
- **Faces in view.** Mostly toward the camera or in a gentle turn. Long stretches in profile or from behind give the AI less to show.
- **Clear light.** Daylight or a well-lit room. Dark scenes, strobes and heavy colored lighting hide features.
- **A calm camera.** Steady shots or smooth moves. Shaky handheld footage and whip pans make faces blur.
- **Few things over the face.** Hands, hair, microphones, glasses and food pass in front of faces in many clips. A moment is fine; a whole scene is harder.
- **Few people.** One to a few main people is ideal. Up to 6 different photos fit into one video.
- **A sharp source.** The original file, not a re-recorded screen or a heavily compressed copy.

## Which clips are harder?

| Easier | Harder |
|---|---|
| One or a few people, close to the camera | A crowd, a stadium, a concert from afar |
| Faces mostly toward the camera | Long shots from behind or in profile |
| Steady or smooth camera | Shaky footage, fast pans, quick cuts |
| Even daylight or studio light | Night, strobes, colored club lights |
| Hands and props away from the face | Eating, drinking, covering the face |
| The key moment within 30 seconds | A story that needs two minutes to land |

A harder clip isn't off-limits. It just may need another take, or a better-chosen part.

## How long should the video be?

You can upload a video 1 to 120 seconds long, up to 150 MB, in any common format. Each new video covers one part of up to 30 seconds, with the original sound.

For an upload, the person search runs on the first 30 seconds automatically, usually in a couple of minutes, and it's free. If the moment you want comes later, choose another part of up to 30 seconds. A good part:

1. Starts close to the action, without a long intro.
2. Shows the people you replace large and well lit.
3. Ends on the punchline, the big move, or the last beat of the music.

## What about people in the background?

In [vswp](/), the people cards show the main people first. People seen briefly or far away are grouped as the background. Without a photo, everyone stays as in the original, so you don't need to fill every card.

Two things to know:

- People who appear only very briefly or are heavily blurred may not be found, so they can't be replaced.
- If you want a whole group in the clip, pick one where everyone is reasonably large; our guide on how to [recreate a trend with friends](/guides/recreate-a-trend-with-friends/) covers casting.

## Should you start from Trends or your own video?

Trends is the quickest start: popular videos where the people are already found, so you only add photos. It's also a good way to see which kinds of clips swap well.

Your own video makes sense when you have a specific scene in mind, such as a family clip, a dance you filmed, or a moment from a party. Film it with the swap in mind: steady phone, good light, faces toward the camera.

## Do you have the right to use the clip?

Use only videos and photos you have the right to use. A video you filmed yourself, with the consent of the people in it, is the safest choice. By pressing "Create video," you confirm you have the right to use the video and the photos. Rules differ by country, and this is not legal advice; see [face swap consent and safety](/guides/face-swap-consent-and-safety/) for more.

## A checklist before you upload

1. The people you replace are large and mostly face the camera.
2. The light is clear and the camera is calm.
3. Hands, props and hair stay off the faces most of the time.
4. The best moment fits into 30 seconds.
5. The file is the original, under 150 MB and up to 120 seconds.
6. You have the right to use the video, and permission for every photo.

If a result still misses, our guide on [what to do when an AI face swap doesn't look like you](/guides/ai-face-swap-does-not-look-like-you/) walks through the fixes.

## FAQ

### Can I use a clip from a movie or music video?

Only if you have the right to use it, both for making the video and for posting it. Rules differ by country, so check before you upload or share.

### Does the video's resolution matter?

A sharper source helps, especially when faces are small. The result comes out at 480p or 720p, and the frame rate follows the source, up to about 30 fps.

### What if my video is longer than two minutes?

Trim it to 120 seconds or less before uploading, then choose the part of up to 30 seconds you want.

### Can I swap someone who is only in the background?

Sometimes. People seen briefly or far away are grouped as the background and can get a photo, but very brief or heavily blurred people may not be found.