Mureka API adds song extension and stems; what Sume Music has

Mureka's API changelog lists song extension, stem separation, remix generation, vocal cloning, lyrics video, music transcription and MIDI and m4a formats, and its latest entry (August 28, 2026) is an enhanced mureka-9.5 model. Sume's Music Router covers one of those jobs, text-to-music with an optional image, and its docs list no extend, stems, remix, transcription or MIDI. For a tool that edits a song after it exists, Mureka's API is the closer fit.

The Mureka facts are from its API Platform changelog, read on 2026-10-02. Sume's are from the Music Router and Music 1.0 docs.

What does Mureka's changelog list?

The changelog runs from January 2025 to August 2026 and groups entries by API updates and model updates. Recent items include a video generation API (August 14, 2026), music transcription (July 9), prompt-to-song without lyrics (June 15) and soundtrack generation for images and videos (May 11). Earlier capabilities it records include vocal cloning, song extension, stem separation, remix generation and lyrics video creation, with parameters for gender control and audio timing, and expanded audio formats including m4a and MIDI. Models named include mureka-9.5, mureka-o2 and the mureka-o1 "first thinking music reasoning model" (July 29, 2025).

The changelog summary does not give prices, song-length limits or request shapes, so none are compared here.

What does the Sume Music Router offer?

POST /v1/music-router/generate takes a prompt of 1 to 5,000 characters, an optional public HTTPS image_url, an optional model, and the standard job fields mode, webhook_url and wait_timeout_seconds. Omit model or send sume/music-auto and Sume picks the engine (Lyria 3.5 today); lyria-3.5 and lyria-3-pro pass through. The result carries an audio artifact, typically audio/mpeg, and result.lyrics holds the model-reported lyrics or section map when present.

Length is steered in the prompt, for example "a 2-minute track" or section markers such as [0:00-0:30] Intro: .... Sending duration or duration_seconds is rejected.

Which jobs map to which API?

What can you do on Sume instead of an edit?

Regenerate with a changed brief, then join or cut audio with the media tools. Timeline audio can concat up to 20 Sume-hosted parts into one gapless file or split a file into ranges, for $0.01 flat per job, so a good intro from one take and an outro from another can be joined without re-synthesis. That is assembly, not editing inside a song.

Every Music Router model charges the same fixed Music price per generation, so trying three briefs costs three generations. There is no seed, which means the same prompt may produce different music; plan on listening.

How does job handling differ for a long song?

A Sume music request is a job. Send Idempotency-Key so a retry does not create a second paid generation, then either poll GET /v1/jobs/{id} or pass webhook_url. With sync mode and wait_timeout_seconds up to 30 you may get the result in the same response, but music often takes longer, so build for the async path first.

The completed job records the model that ran. If you sent sume/music-auto, read the model id from the result instead of assuming it, because the default can change as Sume updates the router. Mureka's changelog does not describe its job model in the entries summarized here, so check its API reference before designing retries.

What should you decide before building on either?

• If you need to extend or separate a finished song, use Mureka's API for that step.
• If you need a bed under a video you already have, use Sume Music plus the timeline soundtrack bed, covered in Sume's timeline docs.
• Check Mureka's own docs for pricing and licence terms; the changelog does not state them.