Lyria 3.5 at $0.08 a song: scoring video on Sume

The Gemini API pricing page lists Lyria 3.5 at $0.08 per song request. Sume does not pass that rate through unchanged: every Music Router model charges one fixed Music price per audio generation, documented as $0.125. To score a video, generate a track through the Music Router and lay it under the footage in Timeline.

What Google lists

From the Gemini API pricing page, read on 2026-10-03: Lyria 3.5 is $0.08 per song request. The Gemini API changelog entry of September 3, 2026 describes full-length song generation with natural vocals and fine-grained duration and structural control.

What Sume charges

The Music Router docs say every router model charges the fixed Music price per audio generation, while the catalog lists the provider list price per model for reference. The Music 1.0 docs state that fixed price as $0.125 per accepted generation, about 56% above Google's $0.08 list figure; confirm the live number with a preview before you submit.

Generate a score

sume/music-auto is the default and resolves to Lyria 3.5 today. Pin lyria-3.5 to name the engine. Duration is not a parameter: duration and duration_seconds are rejected, so steer length in the prompt, for example "a 30-second track" or section markers like [0:00-0:30] Intro: .... Put exclusions in the positive prompt, because a non-empty negative_prompt is unsupported.

curl -X POST https://api.sume.com/v1/music-router/generate \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: score-001" \
  -d '{
    "model": "lyria-3.5",
    "prompt": "Warm lo-fi hip hop, 84 BPM, C minor. Dusty Rhodes chords, brushed drums. A 30-second track. Instrumental, no vocals."
  }'

Lay it under the video

Read the audio artifact from result.artifacts[] where type is audio. Timeline 1.0 takes one audio spine plus ordered video[] slots and returns one MP4. Set audio.duration_seconds to the track length, import both files so their URLs are media.sume.com artifacts, and render with an Idempotency-Key. job.request.routed_model tells you which engine ran.