Dub a 3-minute video into Spanish: a 47.8-cent Sume ledger

Dubbing a 3-minute video into Spanish with Sume jobs costs about $0.478: $0.01 audio detach, $0.03 transcription, $0.12825 text to speech, $0.01 audio join and $0.30 for the render. That assumes 2,700 characters of Spanish script (900 a minute, your own figure to replace) and a translation step you run elsewhere. Sume does not list a translation model in its catalog, so that line is not in this ledger.

The pipeline, step by step

The order follows the audio: pull it from the video, turn it into text with timings, translate, speak the translation, join the takes, and put the new audio on the original picture. Each Sume step is a job on the same envelope and can run async or by webhook, as the jobs and results docs describe.

• Audio detach: POST /v1/audio-detach on the imported video, 16 kHz mono WAV, $0.01 per job.
• Speech to text: POST /v1/stt-1.0/transcribe, $0.01 per audio minute, with sentence segmentation if you want sentence timings.
• Translate the text with your own model or agent. Keep each sentence short enough to fit its original time slot.
• Text to speech: send the Spanish text with language set to es; $0.0475 per 1,000 characters.
• Timeline audio concat: join the takes into one gapless file, $0.01 per job, up to 20 parts.
• Timeline render: video clip plus the new audio, $0.10 per output minute, rounded up.

The ledger

Prices are from the public catalog; character count is an assumption you should replace with the length of your translated script.

Where the ledger is soft

Spanish tends to run longer than English, so the 2,700 characters could be higher and the dub could run past the original. Measure the returned audio before the render; the render price follows the audio length, and a 3 minute 5 second spine bills 4 minutes, which is $0.40. Catalog reserves are also in whole cents, so expect small rounding per job.

The ledger does not make the mouth match. A dubbed track under a talking face is not lip-synced; for on-camera speakers you would need a separate lip-sync step, priced per audio second. For voice-over on screens, products and b-roll it is enough. If the voice and the language disagree, the TTS request returns a mismatch warning that you confirm before it proceeds.

Scaling the ledger

Fixed per-job fees are small, so cost scales with the script and the render. Ten 3-minute videos come to 10 x $0.47825 = $4.7825, about $4.78 before any retries. A single regenerated sentence of 130 characters costs $0.006175 in TTS plus $0.01 for rejoining, which is far cheaper than redoing the 2,700 character script at $0.12825. Keep the per-sentence takes so a single fix never means a full re-run.