YouTube Shorts ads want sound: set generate_audio true on Sume

Google's Shorts ads page says use sound (music, voiceover or both). On Sume, check the model's generate_audio flag in the catalog and set generate_audio true.

4 min readSume
All posts

Google's Shorts ads page tells advertisers to use sound, music, voiceover or both, and to use vertical, sound-on assets. On Sume, send generate_audio true on a model whose catalog entry has generate_audio true, or add a voiceover or music track in a separate step.

What Google's page says

Under creative considerations, the page says to use vertical, sound-on assets that feel social first, and to use sound (music, voiceover, or both) in Shorts ads. In best practices it says that sound has been shown to increase conversions by over 20%. That is Google's claim on its own page, not a Sume number. It also suggests text overlays.

Technical lines on the same page: vertical assets are recommended, video ads can be up to 3 minutes but only the first 60 seconds play in the Shorts feed, and under 60 seconds is recommended.

The Sume side

The video docs define generate_audio as a boolean that tells the model to generate audio with the video, and the default is the audio capability of the model. The catalog descriptor has a generate_audio field that shows whether the model can generate a track. A model that cannot generate audio will not make it just because you ask, so read the field first.

Some models are fixed. MiniMax H3 keeps stereo audio always on with no toggle. sume/auto always has native audio and rejects generate_audio false.

Sound options for a Shorts ad, Sume behavior from the repo docs (read 2026-10-05)
RouteHowNotes
Native model audiogenerate_audio trueOnly where the catalog flag is true
Always-on audioNo toggle (minimax-h3, sume/auto)false is refused on sume/auto
VoiceoverA TTS job, then Timeline audio spineSeparate step
Music bedMusic job, Timeline soundtrackduck_db needs a real spine

A request

The request below asks for a vertical 6-second clip with generated audio. The model is a placeholder for any catalog model that shows generate_audio true and 9:16.

curl -X POST https://api.sume.com/v1/videos \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: shorts-sound-001" \
  -d '{
    "model": "seedance-2",
    "prompt": "A barista pours oat milk, ambient cafe sound, upbeat",
    "duration": 6,
    "resolution": "720p",
    "aspect_ratio": "9:16",
    "generate_audio": true
  }'

Check that the sound survived

After the job completes, confirm the file has an audio track before you upload it to YouTube, since Google's guidance assumes sound on. Run a quick ffprobe for an audio stream, or use video-inspect for the transcript of a voiced clip.

For long ads, trim to the first 60 seconds with video-trim, because only those play in the Shorts feed.

Sources

Related posts

More in Use cases

All Use cases posts

Written by Sume