The best photo for an AI face swap video

What makes a good photo for a face swap video

Run through this before you upload. It takes a minute and it's the cheapest way to a closer likeness.

1. Just you, looking at the camera

   One face, straight on, eyes open. If your best photo is a group shot, crop it to you, or tap your face when the site asks.

2. Your face big in the frame

   Head and shoulders at about arm's length, like a normal selfie. A full-length holiday photo leaves too little face to copy.

3. Soft, even light

   Daylight from a window in front of you is ideal. A dark room, harsh shadows or a bright window behind you hide the shape of your face, and a strong shadow can be copied as if it were part of it.

4. Nothing covering your face

   No sunglasses, hands, masks, caps or hair across the eyes. Anything hidden in the photo has to be guessed in the video.

5. Glasses only if you want them in the video

   Your own glasses travel with your face. Wear them in the photo if they're part of how you look and you want them in the clip; take them off if not. The performer's sunglasses go either way.

6. Hair and facial hair the way you want them

   Your hair, hairline and facial hair come from the photo, so pick one where your beard, stubble or fringe looks the way you'd like to see it on screen.

7. The original, not a filtered copy

   Use the photo from your camera roll, not a beauty-filtered, retouched or AI-enhanced version. Filters smooth skin and reshape eyes and noses, enhancers tend to pull a face toward an average one, and the swap copies whatever it's given.

8. Sharp and full size

   A sharp photo straight from the camera beats a screenshot or a picture saved from a chat app, which are often shrunk and compressed. Our check marks down anything under 512 pixels on its short side; a phone camera photo is far bigger than that.

9. Then add a second angle

   After your first photo, a ＋ angle slot appears. Add one with your head turned a little to the side. You can add up to 3 photos per person, and the best one goes first.

Why the photo matters more in a video than in a photo swap

A photo face swap puts your face onto one still picture. A video remake has to rebuild your face in every frame of someone else's moves: the head turns, the close-ups, the shots from across the room. The photo you add is all it knows about your face, for the free preview and for the video.

That's why the result looks like you rather than a perfect copy. The resemblance is strongest in close, front-on shots and softer in wide or side-on ones, where any face is only a small patch of pixels. A front-facing photo covers the close-ups; a second one, turned a little, helps the shots where your head turns.

Research on AI portraits points the same way. In the PhotoMaker study, likeness improved most going from one input photo to two, and the gains slowed after that. One good photo is enough to start; a second is the cheapest upgrade there is.

• From your photo: face shape, eyes, nose, mouth and skin tone; beard, moustache or stubble (or none); hair and hairline; your own glasses, or none.
• From the clip: body and build, clothes, tattoos and jewellery, the moves, the camera, the light, the setting and the sound.
• Pets are different: the whole animal comes from your photo, so its coat and markings matter as much as its face.

Photo problems: what happens and how to fix them

These are the problems our photo check looks for and the ones we've seen in our own test renders.

The photo check gives advice and still uses your photo; the only photos it refuses are of people who look under 18.

The best photo of your cat or dog

With a pet, the whole animal is swapped, not just the face, so the photo carries more. Use one pet per photo, with its face clearly visible and looking toward the camera, in a shot that shows its coat colour and markings. It doesn't need to be posing: the moves come from the clip, like the two cats standing on their hind legs at the red fence in Smalltown Boy Cats.

For two pets, use two clearly different photos. Pets with similar coats can blend into each other, so choose shots where each one's markings are easy to tell apart. It doesn't have to be two cats: a cat and a dog work too.

Our face detector is built for human faces and skips pet photos, so give yours a quick look yourself: sharp, well lit, one animal. A pet version of Bulin 47 is still in testing.

Photo notes for each trend

Every trend page shows its own photo tips next to the upload. These are the ones that change which photo you should pick.

1. Bulin 47↗ Rising ≈380M views · 1 person · 15s One person, front-facing. Short or tied-back hair reads best under the red light. Your face is a medium close-up for the first 8 seconds, then smaller and partly behind the man filming. See yourself in it
2. Rumpelstiltskin Dance↗ Rising ≈57M views · 1–2 people · 9s A big grin in your photo helps: the imp grins straight at the camera in an extreme close-up. With a friend, photo 1 is the imp and photo 2 the maiden, whose face shows only in two short close-ups. See yourself in it
3. Raindance ≈200M views · 2 people · 28s One photo per person; natural daylight suits the golden-hour light. Photo 1 sits side-on on the pier and turns to look back, so a second, slightly turned photo pays off there. No need to wear green: the outfits stay. See yourself in it
4. Alors on Danse↗ Rising ≈14M views · 2 people · 18s Photo 1 plays the keyboard on the left. Photo 2 leans right into the lens grinning after the cut, so that face ends up very close and partly out of frame. Before the cut both faces are medium-small in a two-shot, so a well-lit, front-on photo matters. See yourself in it
5. Smalltown Boy Cats↗ Rising ≈46M views · 2 pets · 21s One pet per photo, face toward the camera, coat and markings visible. Photo 1 screams up front first; photo 2 dances at the fence first, then they swap places. See yourself in it

What gets changed after the first preview

Under the free preview there's a box where you type a change, and the preview is made again from your original photos. Our first learning report (14 Oct 2026) is a small, early sample that includes our own test runs: 12 previews, all on Bulin 47, and 9 of them were changed. The requests show what a photo can and can't decide:

• “Remove the glasses”, 4 times. Glasses now come only from your photo, and the performer's sunglasses go by default. If you don't want glasses in your video, use a photo without them.
• “Make him slimmer”, 3 times, and “make him taller” once. Build and height come from the clip in a face swap: your face goes onto the performer's body, so a different photo won't change them.
• Removing tattoos, once. They were the performer's. Tattoos, clothes and jewellery come from the clip too, so a different photo won't remove them.

So: get four things right in the photo

Your face, your hair, your facial hair and your glasses (or none) are what the photo decides. Get them the way you want them before you upload and there's less to fix after the preview, where you have 3 changes per trend. Everything else, from the outfit to the dance, is the trend itself.

The preview is a still of you in the scene and takes under a minute. If you like it, the full video is 100 credits, whether it's one person, two people or your pets, with packs from $9.90 for 100 credits and no subscription needed. You pay only when it's ready: it takes about 5–10 minutes, we email you, and if it fails your credits come back automatically.