Face swap for a holiday clip: 4 to 15 seconds with audible speech

Sume's Avatar Face Swap is a Beta endpoint that applies a ready avatar face to a short public source video. The required fields are avatar_handle, video_url and quality (standard, plus or max); there is no default for quality. The docs say the current plan is a source of approximately 4 to 15 seconds with usable audio. See the face swap docs before building on it, because it is Beta.

For a holiday campaign that means one short creator-style beat per call, not a long spot. Use only avatars you own or have consent for, and source footage you have rights to.

Pre-flight checklist

Most failures come from the input, not the model. Walk through these checks before you submit.

The call

It is a model run at POST /v1/models/sume/avatar-face-swap/v1.0/runs. Add an Idempotency-Key. The docs give the usual communication options (async by default, a webhook, or a wait). Poll GET /v1/jobs/:id/status and read /result; when ready, the completed resource exposes a video_url on media.sume.com, and resource_status is the primary readiness signal.

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: holiday-swap-001" \
  -d '{
    "avatar_handle": "your-ready-avatar",
    "video_url": "https://example.com/source-clip.mp4",
    "quality": "standard"
  }'

What to do with the result

Treat the output as a draft. Watch it on a phone, with sound, for lip and face consistency, then trim or caption it for the platform. A swapped clip is a good fit for a localized or personalized variant of a hook, and a poor fit for a clip where the face must match a real person's likeness without their consent.

If you need a scripted talking presenter instead of a swap on an existing clip, use the Avatar Video endpoint, which takes a script, not a source video.

• Beta: expect the 4 to 15 s window to move.
• No prompts or aspect ratio on this endpoint.
• Consent for every face, every time.

Where it fits in a holiday workflow

Face swap is useful for one narrow job: you have a short, approved clip with a performer, and you want the same performance on a ready avatar identity you control. A typical holiday use is a set of hook variants for different audiences, each a 4 to 15 second beat, where the avatar identity stays consistent across the whole campaign.

Plan the batch around the Beta limits. Cut your source clips to the window first with a trim job at $0.02 each, make sure each has clear audible speech, and host them at public HTTPS addresses that Sume's worker can fetch. Signed or private URLs are rejected, so a presigned storage link will not work. Keep an eye on the endpoint's docs, because Beta behaviour and the supported duration can change.

Finally, write down consent. For each avatar, record who agreed to the use, for which campaign and for how long. A swap that is technically valid can still be a problem if a performer or an avatar owner did not agree. Save that record beside the job id and the output URL.