Lyric video captions for an AI song: pass the lyrics as script_text
Lyria 3.5 returns lyrics with the audio. Sume's captions API can take them as script_text to caption a video that carries the song. Steps and limits.
For lyric video captions, generate the song, render it into a video, and call the Sume captions API with the lyrics as script_text. Google's Gemini post, read 2026-10-01, describes Lyria 3.5 for tracks with vocals and lyrics; Sume's Music Router returns them in result.lyrics when the model reports them.
What are the steps?
Generate with POST /v1/music-router/generate. Render an MP4 with Timeline from the audio and a still. Then POST /v1/video-captions with video_url, a style, and script_text. Captions cost $0.20 per job for videos up to 60 seconds.
What are the limits?
result.lyrics is the model's report, not a transcript of the audio. Sume captions need audible speech; the docs name caption_no_speech for silent clips. Sung words may not be treated like speech, so test a sample, and if it fails supply cues with text and timing yourself.
Can I keep my styling?
Yes: style, font and design overrides exist, and source_caption_id restyles an earlier caption. There is no SRT upload.
Which endpoint does each step use?
The steps above, in one place.
| Step | Endpoint or field | Note |
|---|---|---|
| Make the song | POST /v1/music-router/generate | Lyrics in result.lyrics when reported |
| Render video | Timeline | Audio plus a still, MP4 output |
| Caption | POST /v1/video-captions | video_url, style, script_text |
| Caption price | $0.20 per job | Videos up to 60 seconds |
Sources
Related posts
More in Media tools
- Microsoft Advertising image max 2592x2048: send JPEG, not PNG
Microsoft accepts GIF, JPEG and PNG up to 2592x2048 but converts PNG to JPEG and drops animation. Request jpeg from Sume and skip a lossy surprise.
- Microsoft Advertising image sizes: 703x368, 1:1 and 4:1 from AI
Microsoft's Image object has six subtypes from 4:1 to 16:9. Sume cannot set exact pixels, so request the ratio, then crop each subtype to its minimum.
- Nano Banana transparent background: no alpha option, use RMBG
Google's Gemini image docs do not list a transparency option, and Sume's background parameter is for GPT Image 2.5. Cut Nano Banana output out with RMBG 1.0.
- Nextdoor video ad specs: MP4 or MOV, 500 MB, 2 to 300 seconds
Nextdoor video ads are MOV or MP4 up to 500 MB, 2 to 300 seconds, in 16:9 or 1:1 for the Newsfeed. How to render a fitting clip with Sume.
Written by Sume