Pin lyria-3-pro or leave sume/music-auto? Both cost $0.125

On the Sume Music Router, sume/music-auto, lyria-3.5 and lyria-3-pro all bill $0.125 a generation, so price will not pick the model. Pin one for repeatability.

5 min readSume
All posts

Leave the model as sume/music-auto when you want Sume to choose, and pin lyria-3.5 or lyria-3-pro when you need the same engine across a series. The price does not decide it: the Music Router charges the same fixed $0.125 per generation for all three. What you give up by pinning is the router's freedom to choose; what you gain is that two tracks in a series come from the same model.

What the router accepts

The Music Router takes the same body as Music 1.0 and adds an optional model. Its documented ids are sume/music-auto, lyria-3.5 and lyria-3-pro. Music 1.0 itself resolves through the router, so the older route and the router do not give two prices. Music takes no duration or negative_prompt, as the Music docs say: duration is steered in the prompt, and exclusions go into the positive prompt.

Music Router model ids and price; Sume Music Router docs and catalog read 2026-10-09.
Model idPrice per generationUse it when
sume/music-auto$0.125You want Sume to choose and you will listen to the result
lyria-3.5$0.125You want a fixed engine for a series
lyria-3-pro$0.125You want to compare the Pro model against the others

A test that costs $0.375

Write one brief. Send it three times, once per model id. That is 3 x $0.125 = $0.375. Listen blind and note which you prefer for this kind of brief. Repeat with a second brief and the cost is another $0.375. Two briefs and three engines is $0.75, which is less than one minute of a rendered video, so a decision based on your ear is cheaper than a decision based on guesswork.

A single result is a sample, since the same prompt does not return the same track twice. If a series needs consistency, pinning gives the engine, not the exact audio. To reuse one track, save the file and cut it with Timeline audio.

A note on lyrics

The docs state the provider's lyrics field is model-reported metadata, not an audio measurement. Do not use it to check that a track follows your words. Listen to it, or run the audio through speech to text at $0.01 a minute if you need a transcript.

What to record per track

Save the model id you sent, the full brief, the job id and the date. When you pinned a model, the result lists that model; when you left sume/music-auto, note which engine was used if the result reports it, and do not assume. This record lets you reproduce the setup for the next episode, and it settles arguments about why one track sounds different from another. A series of twelve tracks at $0.125 is $1.50 before any cut.

Sources

Related posts

More in Media tools

All Media tools posts

Written by Sume