How to create an avatar from a video: two ways

To create an avatar from a video, you either give the footage to a tool that trains a digital twin from it, or you take one clear frame of the person and make a photo avatar from that still. Training uses the whole recording (HeyGen also clones the voice from it); the frame route works with any avatar API that takes an image, including Sume's.

This page is about talking avatars of a real person, not cartoon filters. Vendor facts come from HeyGen's and Synthesia's own docs, read on 2026-09-28; Sume facts come from Create new avatar and Sume's code.

How do I train an avatar from my footage?

Upload the recording to a product that builds avatars from video. Each sets its own rules for the footage:

How do I make an avatar from one frame of the video?

Sume's avatar API takes no video: POST /v1/avatar-1.0/generate accepts a prompt, a profile, or a reference image. From a recording, the image is your starting point:

• Pause the video on a frame where the face is sharp, front-facing and evenly lit, with only that person in shot, and export it as an image.
• Put the image at a public HTTPS URL. Localhost, private-network, non-HTTPS and non-image URLs are rejected before generation.
• Send it as a photo input with a handle you will reuse, then poll the job until it is completed.

curl -X POST https://api.sume.com/v1/avatar-1.0/generate \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: avatar-from-frame-001" \
  -d '{
    "avatar_handle": "reference_presenter",
    "input": {
      "type": "photo",
      "image_url": "https://example.com/frame.png"
    }
  }'

What carries over from the video, and what doesn't?

The frame route keeps only the look in that one frame. In Sume's current code:

• The likeness is generated, not copied: the photo is sent to an image model with the prompt "Photo of this person".
• The voice is not yours. Current code renders a sample from the avatar's image and clones that voice in English.
• Nothing else from the recording is used, since the create call has no video input.
• To speak in your own voice, clone it in the Sume app (Assets → Voices), where you upload or record audio, and lip sync the frame to it; clone yourself with AI covers that route.

Can I send the video file to Sume directly?

No. Sume has no public upload route for local files, and its API reference says to prefer public HTTPS media URLs in generation requests. The avatar create call takes an image, not a video, so host the frame yourself, as above.

Creating the avatar costs $0.95 per avatar once, plus a 5.5% agent fee by default. The handle then works on every talking video; create a reusable AI avatar covers the rest of the create call, and talking avatar video API covers the next step.