AI avatar 4K: can an avatar video be 1080p or 4K?

You get a 4K or 1080p AI avatar video by upscaling the finished clip: enlarge it by the target height divided by its current height, ×3 from 720p to 4K (2160p) or ×1.5 from 720p to 1080p. On Sume the talking video renders at 720p, the only documented resolution, and Video Upscale 1.0 takes a factor from 1.1 to 4 on up to 30 seconds of input per job.

The avatar facts come from Generate avatar video, the upscale fields from the Sume API reference, and the joining rules from Timeline 1.0, read on 2026-09-29. Prices come from API pricing. For upscaling in general, with scale factors and FFmpeg, see how to upscale video to 4K.

What resolution does an AI avatar video come out at?

720p. The avatar docs say resolution is currently 720p, and the API reference lists 720p as its only value. You choose the frame shape with aspect_ratio: 1:1, 3:4, 9:16, 4:3 or 16:9, with 9:16 as the default. A script must come to an estimated 4–60 seconds. There is no 1080p or 4K setting on the avatar render itself.

How do I upscale an avatar video to 1080p or 4K?

Send the finished clip to POST /v1/video-upscale-1.0/upscale. video_url must be a public HTTPS URL; completed avatar results include public media.sume.com video files, so the avatar's own URL works. Set the factor from the "p" number you have and the one you want: 1080 ÷ 720 = 1.5, and 2160 ÷ 720 = 3.

curl -X POST https://api.sume.com/v1/video-upscale-1.0/upscale \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: avatar-upscale-part-1" \
  -d '{
    "video_url": "https://media.sume.com/artifacts/artf_demo/talk.mp4",
    "upscale_factor": 1.5,
    "duration_seconds": 30
  }'

What if my avatar video is longer than 30 seconds?

The upscale schema caps duration_seconds at 30, and the price is set on at most 30 seconds of input, so cut longer clips into parts of 30 seconds or less first. The docs don't say what happens to a longer input, so don't send one.

This works for 1080p: a 1080p frame's long edge is 1920 pixels, inside Timeline's cap of 2160 per output edge. A 4K frame's long edge is 3840, so on Sume a 4K avatar video stays one upscaled part of 30 seconds or less; how to upscale video to 4K explains why to upscale last.

• Cut: Video trim takes a start and an end or duration on one media.sume.com clip and returns a new MP4.
• Upscale each part with the same factor.
• Keep the voice: Audio detach pulls the original avatar clip's sound into a WAV file.
• Rejoin: Timeline 1.0 takes that WAV as audio.url and the upscaled parts as video[]. In current code a Timeline render keeps only its audio spine (and an optional soundtrack) and drops each clip's own sound, which is why the voice goes in as audio.url.

How much does a 1080p avatar video cost?

Here is a 60-second, 16:9 video on the standard tier with no product image, delivered at 1080p. In current code trim, audio detach and Timeline reserve the listed price and charge the job's own compute cost, never above it, so those lines are ceilings. Every paid call adds a 5.5% agent fee by default. AI avatar video pricing per minute lists every tier.

What does the upscale not promise?

Three gaps in the documentation are worth knowing before you promise a client a 4K file:

• The API reference lists the factor range, not the output pixel sizes; check the file you get back.
• It doesn't say whether the upscaled file keeps its audio track. Play it before you publish, or rebuild the sound on Timeline as above.
• An upscale enlarges the 720p render; it isn't a fresh 4K render, and the docs make no quality claim for it.