Music API · served now
Full songs — vocals, arrangement and mix — from your lyrics and a one-line musical brief, through the /v1/audio/music endpoint. An AI music generation API served from hardware AxForge owns in Sweden: up to five minutes per track, 32 kHz 16-bit stereo WAV, zero prompt retention.
Specifications
| Model | MiniMax-Music3 — full-song music generation: vocals + arrangement from lyrics |
|---|---|
| API model name | minimax-music3 |
| Endpoint | POST /v1/audio/music |
| Output | Tracks up to 5 minutes, 32 kHz 16-bit stereo WAV — base64 in data[].b64_json |
| Vocals & structure | Sung from the lyrics field, with [verse] / [chorus] structure tags; omit lyrics for an instrumental |
| Parameters | prompt (musical description — genre, BPM, instrumentation), lyrics, duration_s (up to 300), seed |
| Licence | MiniMax model licence — commercial use of the service permitted; products built on it name the model and disclose machine-generated audio (details below) |
| Hardware | NVIDIA DGX Spark (GB10, 128 GB unified memory) — owned and operated by AxForge |
| Region | eu-se-1 · Stockholm, Sweden — pinned with x-axforge-region, echoed on the response |
| Price | €0.09 / track (launch pricing) |
Looking for a self-hosted Suno alternative? This is the shape it takes here: a song generation API on machines we own in Europe — the control teams want from self-hosting, without running the model yourself.
Performance
| Metric | Value | Condition |
|---|---|---|
| Generation time | ~2–3 min | per track, end to end |
Measured on our production DGX Spark node, single-stream, 2026-08. A song takes minutes, not milliseconds — call the endpoint from a background worker with a generous client timeout, not inside a blocking user-facing request.
Quickstart
$ export OPENAI_BASE_URL=https://api.axforge.ai/v1 $ curl -sS --max-time 600 "$OPENAI_BASE_URL/audio/music" \ -H "Authorization: Bearer $AXFORGE_API_KEY" \ -H "x-axforge-region: eu-se-1" \ -H "Content-Type: application/json" \ -d '{"model":"minimax-music3","duration_s":180, "prompt":"Warm indie folk, 92 BPM, acoustic guitar, brushed drums, upright bass", "lyrics":"[verse]\nMorning light on the harbour line\nEngines hum and I feel fine\n[chorus]\nWe are on our way, we are on our way"}'
The response is OpenAI-style JSON: data[0].b64_json holds the WAV. Generation takes ~2–3 minutes, so keep the timeout generous and run the call as an async job.
Use case
You have a 40-second product video and a wish: warm, understated, builds at the end. Send the wish as the prompt — genre, BPM, instrumentation, mood — set duration_s to the length of the edit, and the endpoint returns a WAV that drops straight onto the timeline. No stock-library search, no near-miss compromise: a score fitted to the clip, generated on demand.
This is the plain music endpoint doing the work — there is no separate video-scoring product to adopt, and none is needed.
Licence
The model behind this endpoint is MiniMax-Music3. Its licence attaches two obligations to products built on the model — both are plain honesty, not burden:
| Name the model | Products using this endpoint say the music comes from MiniMax-Music3. This page does that for the API itself; a user-facing product carries it forward — a line in the UI or the credits is enough. |
|---|---|
| Disclose machine generation | Every track from this endpoint is machine-generated audio. We say so plainly here; a user-facing product says so to its users too. |
If your product shows the model name and labels the audio as machine-generated, both obligations are met.
Data & privacy
Prompts, lyrics and generated audio are processed in memory in Sweden — not written to disk, not logged, not retained, and never used to train anything. We keep only request metadata (token counts, timestamps, status) for billing and operations. The full policy is at axforge.ai/privacy.
FAQ
Full songs up to 5 minutes: vocals and arrangement generated from your lyrics, with the style set by a musical description — genre, BPM, instrumentation, mood. Omit the lyrics and you get an instrumental. Output is 32 kHz 16-bit stereo WAV.
You mark the song's structure inside the lyrics field:
[verse] and [chorus] tags divide your lines into
sections, and the model writes the melody and arrangement around the structure
you give it, so choruses come back as the hook.
Yes. Omit the lyrics field and describe what you want in the
prompt — for example "ambient piano, 60 BPM, warm tape texture" — and the track
comes back fully instrumental.
Roughly 2–3 minutes per track, measured on our production DGX Spark node, single-stream, 2026-08. Treat generation as a background job: call the endpoint from a worker with a generous timeout, not inside a blocking user-facing request.
Yes — build paid products on this endpoint. The licence asks two things of your product: name MiniMax-Music3 and label the audio as machine-generated. Rights in machine-generated music vary by jurisdiction, so run your specific use past your counsel.
No. Prompts, lyrics and generated audio are processed in memory in Sweden, never persisted or used for training — see the privacy policy. Only request metadata is kept for billing and operations.