AI avatars from a real video, 1:1 like the person

Most avatar engines re-generate the whole face, and viewers notice the waxy skin and frozen eyes. PitchTwin builds every avatar from real video of the person and keeps that footage, so the result looks like them.

Their skin, hair and light

Texture and the room stay exactly as filmed, in full HD.

Their own expressions

Blinks, head moves and gestures come from the person, not an average face.

The same face every time

Each avatar keeps its engine version, so it never drifts after an update.

What you need to record

A phone is enough. The clip is the avatar, so a good clip gives a good avatar.

  • 10 to 120 seconds of one person facing the camera; 30 seconds to 2 minutes works best
  • Steady, even light on the face and the mouth clearly visible
  • Little head movement and nothing in front of the face
  • Natural speech: with 10+ seconds of it, we clone the voice from the clip too

What we do with the clip

Uploads are normalized automatically: phone rotation is fixed, HEVC is converted to H.264, the long side is capped at 1920 px and the frame rate set to 25 fps. The original voice sample is stored with the avatar, so the voice can always be re-created.

Your users can do it themselves

Inside your product, send the clip your user recorded to POST /avatars, or let them upload it straight to us with a one-time upload URL. In the PitchTwin dashboard you can also upload clips from the browser and test them in the Playground.

Questions about avatar from video.

Can I use a photo instead of a video?

No. PitchTwin avatars are built from video so they keep the person's real look and movement. A short phone video is all it takes.

Does the background have to be plain?

No. The room is kept as filmed. Pick a background you're happy to see in every video made with that avatar.

Can one person have several avatars?

Yes. One avatar is one person in one look, so a different outfit or room is a separate avatar.