AI face swap video cost: the $2.76 to $8.25 ceiling per clip

A Beta face-swap clip on Sume is estimated at $2.76 on standard, $3.68 on plus and $8.25 on max. Those figures are the contract estimate in the catalog: the Avatar Video no-product rate for the tier times the 15-second Beta source-video maximum. So the ceiling for one swap is priced as if the source were a full 15 seconds, and quality is required; there is no default.

Face swap puts a ready avatar's face onto a short public video. It is not script-driven generation. If you want a presenter to say new words, use the talking-video route.

Request and limits

One endpoint, three required fields.

Standard, plus or max?

The only way to find out what the extra spend buys is to run the same clip at two tiers. Because a swap is capped near 15 seconds, the experiment is cheap: standard and max together are $11.01 at the ceiling. Run both on a hard case, such as a face that turns away from camera or a clip with fast motion, and compare the edges around the jaw and hairline.

curl -X POST https://api.sume.com/v1/models/sume/avatar-face-swap/v1.0/runs \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: swap-demo-standard" \
  -d '{
    "avatar_handle": "studio_presenter",
    "video_url": "https://example.com/inputs/source-video.mp4",
    "quality": "standard",
    "mode": "async"
  }'

Use face swap or talking video?

• Face swap: you already have a short performance you want to keep, with its movement and audio.
• Talking video: you have a script and want a presenter to deliver it.
• Neither: you want a long clip. Both stop at short windows (swap about 15 seconds, talking video 60).

Limits and care

This is a Beta surface and its limits can change, so read the OpenAPI before depending on it. Use faces you have the right to use. Platform rules on altered or synthetic people differ, and several platforms ask for a disclosure label on face-swapped content, so check the destination before you publish. Keep the source clip and the avatar handle with the output in your own records, so you can show later what was swapped onto what, and delete test clips you no longer need. A short consent note from the person whose footage you use is cheap to collect and hard to reconstruct afterwards.