Lip sync 30 seconds as three 10-second lines: cost by resolution

Three 10-second lines through MiniMax H3 Max lip sync cost $1.875 at 480p, $3.00 at 768p, and $6.00 at 1080p. The audio window is 5 to 14.8 seconds for each job, so a 30-second script cannot go in one request. Sume avatars are async jobs: you submit a request, get a job id back, and read the finished video later. There is no live video session.

The price is ceil(duration_seconds) times the per-second rate, so a 10.0-second line bills as 10 seconds and a 10.2-second line bills as 11.

The rates and the three-line total

Sume bills H3 Max lip sync at the fal list per second for each resolution times the 1.25 house margin: $0.05, $0.08 and $0.16 become $0.0625, $0.10 and $0.20 per second. The API reserves the amount when it accepts the job, captures it on completion, and refunds it on failure.

Splitting the script

Split the script by sentence, not by time. Generate the speech for each line with text to speech, check its length, and only then submit the lip-sync job. A line that comes out under 5 seconds is rejected with invalid_request, and the API never clamps the value. A line over 14.8 seconds would have its tail silently clipped by the provider, so the API refuses the duration_seconds instead of letting you lose speech.

Send the real measured length as duration_seconds. It is the reservation basis only, and the audio sets the output length. Keep each audio file on a Sume-hosted URL and under 10 MB, as the same preflight as Fabric applies.

Request for one line

curl -X POST https://api.sume.com/v1/minimax/h3-max/lip-sync \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: dialogue-line-1-001" \
  -d '{
    "avatar_handle": "product_host",
    "audio_url": "https://media.sume.com/your-line-1.mp3",
    "duration_seconds": 10,
    "resolution": "768p"
  }'

Joining and what else costs money

The three clips are separate jobs, so you join them yourself or with a Timeline render, which provider-pricing lists at $0.10 per output minute, rounded up. A 30-second joined output is one billable minute. Text to speech has its own per-character price, which is not included in the table above.

Draft at 480p to check the sync and the first frame, then render the finals at 768p. Remember that a rerender is a new job and a new reservation, so a three-line video that you redo twice costs three times the table.

Compare with Fabric on the same lines

Fabric is the default talk model and has no 14.8-second ceiling, so the whole 30-second line could go in one job. The provider-pricing rate for Fabric is $0.10 per second at 480p and $0.1875 at 720p, so 30 seconds is $3.00 at 480p and $5.625 at 720p. H3 Max at 768p is $3.00 for the same 30 seconds, but split in three. The difference is $2.625 in H3 Max's favour at the 720p/768p tier.

Price is not the only factor. A single Fabric job gives one continuous clip with one frame rate and no join. Three H3 Max clips need a join, and the guidance is to keep one lip-sync model for the entire run. If the final video is one speaker saying one long thing, one Fabric job is simpler. If it is three short, separate lines with cuts between them, H3 Max fits well.

If a job fails

The reservation is released on failure, so a failed line does not cost you the line. Retry just that line with the same Idempotency-Key only if the original request itself timed out on the client. If the job reached a terminal failed state, change the cause first (for example the audio host or the length), then submit a new request with a new key. Do not resubmit all three lines because one failed; the other two jobs stand on their own.