Skip to content
Guides

How to take photos that work for two-person AI videos

A two-person template asks the model to do something hard: keep two different faces apart for up to 15 seconds while both of them move. Seedance models take image references together with a text prompt.[1] Clear, well-lit photos give the model the best starting point, and these 6 habits cover what you can check yourself before you spend any credits.

1. One person per photo

Upload exactly one face in each photo. A group shot forces the model to guess which face is yours, and it will sometimes guess wrong or blend two faces into one. Portrait mode keeps the subject in sharp focus and blurs the background.[2] Each photo can be a JPG, PNG or WebP file of up to 5 MB.

Good example
One face, clear background.
Avoid
Two faces in one photo.

2. Make the face big and front-facing

The face should fill about a third of the frame and look at the camera, or very slightly off it. Tiny faces and strong side angles give the model too little to hold on to, and the face drifts once the person starts to move. Pixel Camera’s palm timer requires the face to be fully visible in the frame.[3] Use the original file rather than a screenshot: the photo check rates a shorter side of 800 px or more as Good, gives a warning from 500 to 799 px and flags anything smaller.

Good example
Large, front-facing.
Avoid
Face too small.
Avoid
Strong side angle.

3. Use even light and keep the face clear

Soft daylight from the front is ideal. Sunglasses, hats pulled low, hard shadows and motion blur all remove detail the model needs to recognise you. The photo check marks a photo as Too dark, Dim, Bright or Too bright and tells you how to fix the light.

Good example
Even, soft light.
Avoid
Sunglasses cover the eyes.
Avoid
Hat hides the forehead.
Avoid
Too dark.
Avoid
Soft focus.

4. Wear different colours

If both people wear similar tops, the two photos are harder to tell apart. A different top colour for each person keeps the left and right person easy to recognise.

5. Match the framing

Take both photos at a similar distance. A close-up next to a full-body shot can produce two people at very different scales in the final video, which looks wrong even when the faces are right.

6. Check the free scene preview before you pay

Our photo check scores each photo from 0 to 100 on 4 checks before you spend anything, and the scene preview places your two photos into the template layout. If one photo scores low, swap it and the new photo is checked on the spot. The check and the scene preview are free and use 0 credits, while a video is 5, 10 or 15 seconds long in 720p.

The short checklist

  • One face per photo
  • Face fills about a third of the frame
  • Even light, no sunglasses or hats
  • Different top colours for the two people
  • Same distance in both photos
  • Both photos scored well in the free check

Questions about photos

Do both people need their own photo?

Yes. Upload one face per photo. A group shot makes the model guess which face is yours.

Does the photo check cost credits?

No. The photo check and the scene preview are free. Credits are used only when you generate a video.

How long are my photos kept?

Source photos delete from our storage when your video finishes or fails, with automated cleanup starting 55 minutes after rendering begins. Copies held by our rendering partner follow their own retention rules. Finished videos remain in your account for 7 days.

Which files and sizes does the photo check accept?

JPG, PNG and WebP files of up to 5 MB each. A shorter side of 800 px or more rates Good on resolution, and each photo gets a score from 0 to 100.

Sources

  1. Seedance 2.0seed.bytedance.com

    SupportsSeedance models take image references together with a text prompt.

  2. Use Portrait mode on your iPhone - Apple Supportsupport.apple.com

    SupportsPortrait mode keeps the subject in sharp focus and blurs the background.

  3. Take a selfie on your Pixel phone or tablet - Pixel Camera Helpsupport.google.com

    SupportsPixel Camera’s palm timer requires the face to be fully visible in the frame.

Your turn in the orange room.

Make free scene preview