Face Swap a 15-Second Clip With a Consented Avatar: What It Costs

The short answer

A 15-second face swap on Sume costs $2.76 at standard quality, $3.675 at plus and $8.25 at max (rates $0.184, $0.245 and $0.55 per second, catalog read 2026-10-08), plus $0.95 once to create the avatar. The endpoint is Beta and applies a ready avatar to a public HTTPS source video of roughly 4 to 15 seconds with usable audio.

What the endpoint takes

POST /v1/models/sume/avatar-face-swap/v1.0/runs requires avatar_handle, video_url and quality (standard, plus or max; there is no default). The source must be a fetchable public HTTPS video. Localhost, private network, non-HTTPS, signed or private URLs and provider task URLs are rejected. By design it does not take prompts, transcripts, duration, aspect ratio, avatar ids in the body or provider fields.

Consent and fit

The use case is a company that has its own spokesperson, or a creator who made an avatar of themselves, and wants to put that face into a clip they already shot, for example the same demo performed by a different person. The avatar owner must agree to each use. Do not use another person's face, a public figure's face or a customer's face without written permission, and tell viewers when a face is synthetic where the platform or the law asks you to.

If the goal is a new talking video rather than a swap on an existing clip, avatar videos is the documented route. Use face swap only when the performance, background and timing in the source are already right.

Plan the test ladder before you spend. A first pass at standard quality on the shortest cut of the scene costs $0.736 for 4 seconds. If the identity holds and the lip movement matches the audio, repeat at plus ($0.98 for the same cut) and compare. Only then run the 15-second source, and reserve max quality for the final, where 15 seconds is $8.25. That ladder costs $9.966 in total, against $24.750 for three blind attempts at max. Face swap is a Beta endpoint, so check the docs page for changes before each campaign, and keep the unedited source clip, the avatar consent record and the job id together in the project file.

• Run one 4-second test at standard quality: $0.736.
• Source audio is kept as the usable track; the docs ask for usable audio in the source.
• Poll with GET /v1/jobs/:id/status and read the Sume video_url from the result; store it, not a provider URL.