YouTube Audio-to-Short: 8 s to 3 min, English transcript required

YouTube's Audio-to-Short converts an audio-first video that has a transcript into a Short between 8 seconds and 3 minutes long. YouTube's help page lists it as desktop only, as needing an English transcript, as limited to Australia, Canada, the UK and the US, and as asking you to acknowledge AI disclosure. Sume does not drive that tool. What you can do with the API is check that your source fits the length window and produce a clean transcript to attach.

What the help page lists

The tool sits in Create content for Shorts using AI-generated features, read 2026-10-04. On that page Audio-to-Short is the feature that asks you to read the AI disclosure and privacy notice and click Continue; for the other AI tools it says the tool discloses AI use automatically. YouTube's September 2025 blog post (Unpacking the magic of our new creative tools) places the broader family of tools, including Edit with AI and Speech to Song, under SynthID watermarks and content labels.

Check the length window

The page gives 8 seconds to 3 minutes as the length of the Shorts you create and states no maximum input length, but checking your clip's length is still a cheap sanity step for a podcast clip you trimmed by eye. Run video inspect with frames: false to get the probe: duration_seconds, has_audio, and the container details. Inspect reads one clip already on media.sume.com (up to 1800 s), so import first with POST /v1/media-imports.

If the clip is too long, video trim cuts a [start, end) range into a new MP4 for a flat $0.02 per job per the docs; the output must be between 0.2 and 900 seconds.

Make the English transcript

Set transcribe: true on the inspect call and Sume STT 1.0 returns transcript.text, word timings, and with segmentation.mode: "sentence" gapless sentence segments. The public rate in the docs is $0.01 per audio minute; omit duration_seconds and the hold reserves one minute. Pass language_code: "en" as a hint if the audio is English, or omit it to auto-detect. A clip with no audio track fails as inspect_source_has_no_audio, so check probe.has_audio first.

This produces text you own; it does not upload anything to YouTube or feed Audio-to-Short for you. Whether the Studio tool accepts a transcript you bring, or builds its own, is not stated on the help page, so treat the transcript as a reference copy for review.

curl -X POST https://api.sume.com/v1/video-inspect \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: a2s-check-001" \
  -d '{
    "video_url": "https://media.sume.com/artifacts/artf_demo/podcast.mp4",
    "frames": false,
    "transcribe": true,
    "language_code": "en",
    "duration_seconds": 180
  }'

Burned-in captions are a separate job

If you want captions drawn into the pixels rather than YouTube's own, video captions burns them onto a public HTTPS URL for $0.20 per job on clips up to 60 seconds under the docs' fixed estimate. For clips longer than 60 seconds, confirm the price in GET /v1/catalog before you queue a batch.