Short answer
Use one sharp, front-facing photo of one adult, with the head and shoulders taking up a good part of the frame and even light from the front. The face is roughly 25 pixels wide at the start of the clip and under 10 by the end, so a soft, distant or filtered photo is the single most common reason a render comes back looking like the original rider.
Try it with one photoFlamingo Float AI photo guide
One seat, one face, eleven seconds of shrinking frame — the photo does more work here than on any other effect.
Updated 2026-11-04
The short checklist
- One person per photo. This template has one seat, and two faces in one frame fuse into one.
- Head and shoulders filling a good part of the frame — closer than you would use for the band template.
- Face square to the lens, eyes open and visible, no sunglasses.
- Even light from the front: a window or a lamp in front of you, never behind.
- Sharp. A phone photo is fine; a screenshot of a phone photo usually is not.
Why the face is the only thing that matters
In this framing the person on the float is not the subject of the shot — the drop and the float are. The head is about 25 pixels wide in the opening seconds, and by the last frame the float is a pink shape at the far end of the pool with a few pixels of face on it.
The model only has those pixels to work with. Everything else it knows about you comes from the photo, so the photo has to carry the likeness on its own: a large, sharp, evenly lit face gives it something to shrink; a small or soft one gives it nothing, and it falls back on the reference rider's face.
What actually breaks
| Photo | What happens |
|---|---|
| A face too small in frame | The render keeps the original rider's features, because there is not enough identity to copy |
| Sunglasses | The eyes are invented, and the face stops matching the person who uploaded |
| A heavy beauty filter | The filter's smoothing survives the render and the face reads as a stranger |
| Backlight or hard side light | Half the face goes dark and stays dark for all eleven seconds |
| Two people in one frame | One face wins and the other disappears into the float |
| A hat over the hairline | The hairline is guessed, and the new person stops being recognisable from the wide shot |
Framing: closer than you think
If you have used the band template on this site, you already know that a half-body photo works there. Here, go tighter. A head-and-shoulders photo where the face takes up a third of the frame is the right shape; a full-body holiday photo puts the face at the size the render is going to shrink it to anyway, and the likeness collapses.
This is also why the template refuses a second photo. Two faces at that pixel budget would each get roughly 12 pixels at the start and 5 at the end — enough for a shape, not enough for a person.
Clothes, background and light
- Clothes come from your photo. A hoodie or a coat becomes a plain t-shirt in the same colour, because the scene is a pool in the sun.
- Background does not matter. The pool, the float and the skyline are all replaced by the reference.
- Daylight from the front is ideal. The scene is a hazy midday rooftop, so a photo shot in sunlight matches the grade best.
- Avoid a photo shot into the sun: the render inherits the silhouette and the face stays dark.
If the first render keeps the original face
- 1Swap in a sharper, closer photo — the same one, cropped in, often fixes it.
- 2Remove sunglasses, hats and filters from the source before uploading.
- 3Try 720p rather than 480p: more pixels on the face is the whole difference at this framing.
- 4If it still misses, the photo is probably too soft to carry the likeness, and a different photo is the fix.