Mureka V9.5 Music API on WaveSpeedAI: Generate Song, Generate BGM, and Prompt-to-Song

Mureka V9.5 is available on WaveSpeedAI as of September 1, 2026, through three endpoints: mureka-ai/mureka-v9.5/generate-song for lyrics-to-song, mureka-ai/mureka-v9.5/generate-bgm for instrumental background music, and mureka-ai/mureka-v9.5/prompt-to-song for a full song, lyrics included, from a description. Pricing is flat per track: $0.225 for the first two, $0.75 for prompt-to-song. All three return mp3, wav, or flac, and all three run through the official Mureka API.

This post covers what each endpoint does, the exact parameters, what you pay, and how V9.5 differs from the Mureka V9, V8, V7.6, and O2 endpoints that were already in the catalog. Every parameter and price below is from the live WaveSpeedAI model schemas on September 5, 2026.

What Mureka says changed in V9.5

Mureka announced V9.5 on August 31, 2026. The press release frames the release around three things: arrangement logic (songs that hold a structure from intro to outro rather than drifting), vocal performance (phrasing and emotional delivery that sit inside the mix instead of on top of it), and consistency across a full-length track. Mureka says development was informed by more than 25,000 user survey responses, and reports internal test figures of 97.0 percent prompt adherence, 95.7 percent genre matching, and lead vocals rated convincing in 61.0 percent of cases. Those numbers are Mureka’s own, from an unpublished internal evaluation; treat them as the vendor’s claim rather than an independent benchmark.

Coverage in Variety notes that the same launch introduced an O3 reasoning model. O3 is not on WaveSpeedAI at the time of writing; the three endpoints below are all V9.5.

The three endpoints

Endpoint	Input you must supply	What it produces	Price
mureka-ai/mureka-v9.5/generate-song	lyrics	One complete song sung from your lyrics	$0.225 per song
mureka-ai/mureka-v9.5/generate-bgm	prompt	One instrumental track, no vocals	$0.225 per track
mureka-ai/mureka-v9.5/prompt-to-song	nothing required; prompt or styles in practice	One complete song with model-written lyrics	$0.75 per song

None of the three has a pricing formula that varies by parameter. Whatever you pass for output_format, references, or styles, the price is the flat figure above, and each request yields exactly one track.

generate-song: lyrics in, song out

Use this when you already have the words. The model writes melody, arrangement, and a vocal performance around them.

Parameter	Required	Type	Notes
lyrics	yes	string	Up to 5,000 characters. Section markers such as verse, chorus, and bridge help the arrangement.
prompt	no	string	Style direction, up to 1,024 characters: genre, mood, instrumentation, tempo, vocal style.
output_format	no	mp3, wav, flac	Default mp3.
reference_id	no	string	Mureka reference file ID that guides style or arrangement.
vocal_id	no	string	Mureka vocal file ID that guides vocal identity.
melody_id	no	string	Mureka melody file ID that guides the melodic contour.

reference_id and melody_id are produced by mureka-ai/create-upload-id; vocal_id comes from mureka-ai/vocal-clone. Those are separate, separately priced calls, so a vocal-guided pipeline is two requests, not one.

generate-bgm: instrumental from a prompt

The smallest interface of the three. There are no lyrics, no vocals, and no reference IDs.

Parameter	Required	Type	Notes
prompt	yes	string	Up to 1,024 characters. Describe mood, genre, instruments, tempo, and the scene the music sits under.
output_format	no	mp3, wav, flac	Default mp3.

Keep the prompt about the music, not about words. The model is producing a bed for video, games, ads, and podcasts, and a prompt that reads like lyrics tends to fight that.

prompt-to-song: the model writes the lyrics

This is the endpoint for a full song from a one-line brief. The schema has no required fields at all; in practice you should pass at least one of prompt, styles, reference_id, or vocal_id so the model has something to work from.

Parameter	Required	Type	Notes
prompt	no	string	Up to 2,000 characters: theme, genre, mood, instruments, tempo, vocal direction.
styles	no	array of strings	Genre tags such as pop, rock, jazz, r&b, edm, lo-fi.
output_format	no	mp3, wav, flac	Default mp3.
reference_id	no	string	Mureka reference file ID.
vocal_id	no	string	Mureka vocal file ID.

At $0.75 it is 3.3 times the price of generate-song. You are paying for the lyric writing and arrangement decisions in one pass. If you have lyrics, or can generate them cheaply with mureka-ai/generate-lyrics at $0.015 per call, generate-song plus your own lyrics is the cheaper route to the same kind of output.

Output format and delivery

All three endpoints return one audio file per request. output_format chooses mp3 (default, smallest), wav, or flac. The file is returned as a URL in the prediction’s outputs list, following the same asynchronous submit-and-poll pattern as every other WaveSpeedAI model: POST https://api.wavespeed.ai/api/v3/<model_uuid>, then GET https://api.wavespeed.ai/api/v3/predictions/<id>/result until data.status is completed.

A minimal generate-bgm request:

{
  "prompt": "warm lo-fi hip hop, mellow Rhodes piano, vinyl crackle, 80 bpm, late-night study vibe",
  "output_format": "wav"
}

How V9.5 compares with the earlier Mureka endpoints

The catalog still carries the previous generations, and they are not being removed. The practical differences:

Endpoint	Price per track	Tracks per request	Lyrics limit	Extra references
mureka-ai/mureka-v9.5/generate-song	$0.225	1	5,000 chars	reference, vocal, melody
mureka-ai/mureka-v9/generate-song	$0.07	1 to 3	3,000 chars	reference, vocal, melody
mureka-ai/mureka-v8/generate-song	$0.07	1 to 3	3,000 chars	reference, vocal, melody
mureka-ai/mureka-o2/generate-song	$0.07	1 to 3	3,000 chars	reference, vocal, melody
mureka-ai/mureka-v7.6/generate-song	$0.045	1 to 3	3,000 chars	reference, vocal, melody
mureka-ai/mureka-v9.5/generate-bgm	$0.225	1	n/a	none
mureka-ai/mureka-v9/generate-bgm	$0.07	1 to 3	n/a	instrumental
mureka-ai/mureka-v8/generate-bgm	$0.07	1 to 3	n/a	instrumental
mureka-ai/mureka-v7.6/generate-bgm	$0.045	1 to 3	n/a	instrumental
mureka-ai/mureka-v9.5/prompt-to-song	$0.75	1	model-written	reference, vocal

Three things stand out.

Price. V9.5 is 3.2 times the per-track price of V9 for song and BGM. The earlier endpoints bill base_price * number_of_songs, so three V9 variations cost $0.21, less than one V9.5 song.

One track per request. The number_of_songs parameter (1 to 3) is gone in V9.5. If your pipeline generated three candidates and let a user pick, you now make three calls.

Longer lyrics. The lyrics ceiling rises from 3,000 to 5,000 characters, which matters for songs with several verses or a spoken bridge.

There is also a gap: V9.5 has no extend endpoint yet. Extension is still served by mureka-ai/mureka-v8/extend-song at $0.15 per call and by the V7.6 equivalent. Prompt-to-song is entirely new; none of the earlier generations offered a lyrics-free song endpoint.

Which one to use

• You have lyrics and care about vocal quality: mureka-ai/mureka-v9.5/generate-song. This is where Mureka’s V9.5 claims about vocals and arrangement apply.
• You need a music bed under a video, ad, or game scene: mureka-ai/mureka-v9.5/generate-bgm. No vocals to manage and the cheapest V9.5 entry point.
• You have a theme and no words, and want one call: mureka-ai/mureka-v9.5/prompt-to-song.
• You are generating many candidates or working at high volume: mureka-ai/mureka-v9/generate-song or generate-bgm at $0.07 with up to three tracks per request. Run V9.5 on the shortlist.

For rights questions on generated music, Can you use AI music generation API outputs commercially? covers the general picture. For a wider survey of what is in the catalog, see Best AI music generators in 2026.

Try each endpoint from its model page: generate-song, generate-bgm, and prompt-to-song. One API key covers all three and the earlier Mureka generations.