APIText to Speech API

Give the video a voice
and a soundtrack, from one key

One call speaks a line of your timeline and comes back word-aligned by Whisper, so subtitles land on the syllable. Another writes the song under it. Four voice providers, one endpoint each.

POSThttps://autostud.ai/api/v1/actions/timeline.generate_audio

Get an API key API reference

Three calls, silence to soundtrack

Voice the block, add a song if you want one, then render the whole thing.

Voice a block

Point the call at a block of the timeline and the voice you want. What comes back is the audio, its exact duration, and the Whisper word timings that make subtitles land on the syllable.

export AUTOSTUD_API_KEY="sk_live_xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx"

curl -X POST https://autostud.ai/api/v1/actions/timeline.generate_audio \
  -H "Authorization: Bearer $AUTOSTUD_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "video_id": "0f4c8b71-2d95-47a3-9e6c-15b8ad7f3c20",
    "block_id": "block_2",
    "voice_settings": {
      "voice_provider": "elevenlabs",
      "voice_id": "rachel",
      "playback_speed": 1
    }
  }'

Write the song

A brief, a style or your own lyrics. Songs are asynchronous: the call returns a task and music.completed brings the hosted tracks back.

curl -X POST https://autostud.ai/api/v1/actions/music.generate \
  -H "Authorization: Bearer $AUTOSTUD_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "title": "Lighthouse",
    "style": "warm lo-fi, slow, no drums",
    "prompt": "A calm bed under a voice-over about the sea",
    "instrumental": true
  }'

# Already wrote the words? Send lyrics instead of prompt.

Render it together

Audio decides the length of a block, so once the voice is in, renders.create turns the timeline into an mp4 with everything in sync.

curl -X POST https://autostud.ai/api/v1/webhooks \
  -H "Authorization: Bearer $AUTOSTUD_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "url": "https://your-app.com/hooks/autostud",
    "events": ["music.completed", "music.failed"],
    "description": "Audio"
  }'

What you can pass

Three fields to speak a block, one to brief a song.

video_id

The video whose timeline holds the block you are voicing.

block_id

Which block gets the voice. Its duration follows the audio, not the other way round.

voice_settings

The voice itself: provider, voice reference, speed. Your own provider key means the generation costs nothing.

prompt

For music.generate: the brief for the song. Pass lyrics instead when you already wrote them.

Four providers, one shape of call

Voices are a catalog, not a hardcoded list. Bring your own provider key and generations are free; use ours and they are billed in credits like everything else.

ElevenLabsGoogleCartesiaFish Audio

Songs report themselves

Speech comes back in the response. A song takes longer, so it fires an event instead: register a webhook once and the tracks arrive hosted, with the credits already reconciled.

generation.completedmusic.completedmusic.failed

Questions developers ask first

Do I get word-level timings?

Yes. Whisper alignment runs on what was just generated, so the response carries per-word timings at full precision, never rounded to the nearest second.

Can I use my own ElevenLabs key?

Yes, and it is the point of the dual mode: with your key the generation costs zero credits and the audio is still hosted, aligned and dropped onto the block for you.

How long does a song take?

Minutes, not milliseconds, which is why music is asynchronous. Read it back with music.get_task if a webhook is not an option in your stack.

What happens to the block duration?

The audio decides it. A block is as long as the speech inside it plus the gaps you configured, so nothing has to be timed by hand after the fact.

OpenAPI spec Error codes Explore the API