Talking head video format: aspect ratio, size, and file

A talking head video is one person speaking to camera, framed from about the chest up, and its format follows where it will play: vertical for short-form feeds, 16:9 for a standard YouTube video, and square or 4:5 for some social feeds. Sume's avatar videos come in five shapes, 9:16 (the default), 3:4, 1:1, 4:3, and 16:9, as MP4 files at a documented 720p.

Platform guidance is quoted from each platform's own page, read on 2026-09-27. Sume facts come from Generate avatar video and the avatar video schema in the Sume API reference; anything called current behavior is read from Sume's code.

Should a talking head video be vertical or horizontal?

Pick the shape from the destination. YouTube treats square or vertical uploads of up to 3 minutes as Shorts and says to use a wider ratio such as 16:9 for long-form videos. The rows below quote each platform's own page, four of them ad specs; Meta video ad specs covers safe zones.

Which formats do Sume avatar videos come in?

Two fields on POST /v1/avatar-1.0/talking-video set the frame, and the file type is fixed:

• aspect_ratio: 1:1, 3:4, 9:16, 4:3, or 16:9. The default is 9:16, the vertical shape of a UGC-style clip; the docs' own example pairs it with a casual bedroom scene, as below.
• resolution: currently 720p, the documented resolution. There is no width or height field.
• The file: the final video is an MP4, and in current code its audio is muxed as AAC.

curl -X POST https://api.sume.com/v1/avatar-1.0/talking-video \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: ugc-vertical-001" \
  -d '{
    "avatar_handle": "product_host",
    "script": "Okay, three things I wish I knew before I started running.",
    "scene": { "type": "prompt", "prompt": "Casual bedroom framing, native UGC lighting" },
    "aspect_ratio": "9:16"
  }'

What happens if I send 4:5 or another ratio?

You get a 9:16 video, not an error. The schema says values outside the supported set are treated as the default, and today the code falls back to 9:16 for any unsupported string. Check the ratio before you send it.

For 4:5, render 3:4, the closest shape, and crop a thin band off the top and bottom with video filter. Crop takes fractions of the frame; keeping the full width and 0.9375 of the height turns 3:4 into 4:5 (our arithmetic: 0.75 ÷ 0.9375 = 0.8). The source must be a media.sume.com clip in your workspace, such as the avatar output, and the audio carries over.

curl -X POST https://api.sume.com/v1/video-filter \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: crop-4x5-001" \
  -d '{
    "video_url": "https://media.sume.com/artifacts/artf_demo/talk-3x4.mp4",
    "ops": [{ "op": "crop", "x": 0, "y": 0.03125, "width": 1, "height": 0.9375 }]
  }'

What changes when I join clips into a longer video?

A Timeline 1.0 render sets its own frame. It outputs 1080×1920 by default, so a 16:9 talking head joined without an output comes out vertical. Set output.width and output.height (even integers from 256 to 2160) to the shape you rendered, and each slot's fit decides how the clip fills that frame: cover (the default), contain, stretch, or blur. Convert a landscape video to vertical compares the fits.

What can't I set?

• A ratio outside the five, such as 4:5 or 21:9. Render the nearest supported ratio and crop it with video filter.
• A pixel size or a resolution other than 720p on the avatar video itself.
• More than one shape per job: aspect_ratio is one top-level field, so render one video per shape you need.