AI Music
Bring a song, vocal, or instrumental you already love. Give it new lyrics. AresGen renders a new track that keeps the original sonic identity, powered by MiniMax Music-01.
One engine · Three modes · MP3 output
Routed engine
One tuned engine, three honest modes (song, voice, and instrumental), all reference-conditioned.
By the numbers
Backend surfaces
Song, voice, instrumental: each mode routes through the same reference-conditioned engine. No vendor drift between renders.
Upload a reference song. Provide new lyrics. AresGen re-renders the track with the new lyrics while preserving the genre, tempo, and overall sonic identity of your reference.
Upload a reference vocal track. Provide new lyrics. AresGen generates a new vocal performance in the same vocal character, ready to drop into an existing instrumental.
Upload a reference instrumental. Provide lyrics. AresGen layers a generated vocal over the instrumental, matched to the tempo and key of the reference.
One platform
Single provider, intentional. No multi-engine drift between renders.
Capabilities
Character preservation applies to surfaces with a vocal (Song and Voice), not Instrumental. Every other capability is enumerated below from the same facts module the audit gate enforces.
| Capability | Song | Voice | Instrumental |
|---|---|---|---|
| Reference-conditioned | ● | ● | ● |
| Custom lyrics | ● | ● | ● |
| Audio file upload (mp3/wav/ogg) | ● | ● | ● |
| Reference via URL | ● | ● | ● |
| MP3 download | ● | ● | ● |
| Vocal character preserve | ● | ● | — |
How a render runs
Click each step to see exactly what AresGen sends to MiniMax Music-01. No auto-advance. The chip you tap is the step you see.
Step: Upload reference
Drop in an mp3, wav, or ogg up to 10 MB, or paste a public audio URL. The reference is uploaded to the aimlapi.com purpose-tagged endpoint as either voice or instrumental input.
Pricing
Music renders are metered at 0.05 credits per generation, included in every paid AresGen plan. Free trial accounts start with enough credits for a handful of test renders. No separate music subscription.
Use cases
Upload your brand jingle in voice mode and supply the new campaign lyrics. The vocal character carries over while the message updates.
Use song mode to test different lyric drafts against your demo. Same arrangement, different message, every render.
Drop a loop or instrumental into instrumental mode and supply the lyrics. Output is a mixed MP3 with the new vocal sitting on the loop.
Upload a vocal-only reference in voice mode once, then render different lyrics against it for episodic content.
FAQ
AiMusic accepts mp3, wav, or ogg files up to 10 MB. You can also pass a public audio URL, and AresGen will fetch the file and forward it to the model.
Lyrics are passed through to MiniMax Music-01 as-is. The model supports several major languages including English, Mandarin, Japanese, Korean, Spanish, French, and German. Results vary by genre and language combination; English is consistently strongest.
You do. AresGen does not store, redistribute, or grant licensing over your uploaded reference. You are responsible for having the rights to use any audio you upload: original recordings, properly licensed samples, or material you have explicit clearance for.
A single MP3 file, decoded from the model response and saved to your music library. The output is a single mixed track. AresGen does not return separate vocal, drum, or instrumental layers; the model does not produce them.
AresGen does not grant or restrict commercial rights beyond what aimlapi.com and MiniMax specify in their terms. Review the current aimlapi.com terms and any applicable MiniMax licensing before using output in paid campaigns or broadcast. For sensitive campaigns, confirm with the provider directly.
AresGen routes AI Music through MiniMax Music-01 because the surface is reference-conditioned, not prompt-to-song. The honest job-to-be-done is: bring a song, vocal, or instrumental you already like, give it new lyrics, get back a track that keeps the original sonic identity. Multi-engine orchestration here would mean tone drift between providers: the opposite of what reference-conditioned regeneration promises. One tuned engine, three honest modes, predictable output.
Pair this with
