S
Sonauto — AI Music Generator ◈ MELODIA · UNLIMITED · COMMERCIAL
Try
◈ MELODIA · v3 · LATENT DIFFUSION

Text into song. Unlimited. Yours to keep.

Sonauto turns a text prompt into a complete song using Melodia — a latent-diffusion model built for continuous-space control instead of discrete tokens. Free unlimited generation, real commercial rights on every tier, inpainting, stem export, BPM control. Rebranded to Treblo in 2026 — same team, same songs.

Try Sonauto → See how Melodia works
∞
Free generations
All
Commercial rights
4:45
Track length · v3
Stems
Vocals · drums · bass · instr
Melodia · v3
GENERATING
synthwave sunset drive, arpeggiated synths, 128 bpm, A minor
WAVEFORM · GENERATED TRACK 03:24
STEMS · EXPORT-READY 4/4 SEPARATED
Vocals
Drums
Bass
Instr
BPM
128
Key
A min
Length
3:24
Model
v3
GENERATED IN 74s · READY
Commercial: yes
Latent diffusion
·
Melodia · v3
·
Unlimited free
·
Full commercial rights
·
Inpainting
·
Stem export
·
BPM + key control
·
Y Combinator backed
·
Latent diffusion
·
Melodia · v3
·
Unlimited free
·
Full commercial rights
·
Inpainting
·
Stem export
How it works

Prompt. Diffuse. Edit.

Three moves from blank canvas to shippable track. Latent diffusion in the middle does the heavy lifting — you drive at the ends.

1 Prompt

Describe the vibe.

Type a text prompt describing style, mood, tempo. Add tags for instruments and genre. Optionally supply your own lyrics or a MIDI file to guide chord progression and melodic contour.

› synthwave sunset drive, 128bpm
2 Diffuse

Melodia generates.

The Melodia model runs diffusion in continuous latent space. Noise transforms into structured audio over the denoising steps. Streaming starts fast; full track lands under 90 seconds typical.

NOISE → SIGNAL
3 Edit

Inpaint, extend, export.

Chorus perfect but verse off? Inpaint just that section. Need it longer? Extend 60-80s. Want stems? Export vocals, drums, bass, and instrumental separately. Own the file. Ship the track.

INPAINT
The Melodia model

Continuous latent space beats discrete tokens.

Most AI music models predict audio the way language models predict text — token by token. Melodia runs diffusion in continuous latent space instead, the same class of technology as Midjourney but tuned for full-length song structure.

Latent diffusion · continuous control

Diffusion starts from noise and denoises toward a structured signal over many steps. Because the space is continuous, you can nudge parameters mid-generation — BPM, key, structure — rather than being locked into a fixed token vocabulary. The founders published a research paper documenting coherent tracks up to 4:45 with this architecture.

Inpainting · surgical edits

Every other major AI music tool makes you regenerate the whole song if one section is off. Melodia's inpainting isolates a region — the second verse, the bridge, a specific measure — and regenerates just that region while preserving the surrounding audio exactly. Producers cite this as the single reason they switched.

Stems + BPM + key · producer surface

Export vocals, drums, bass, and instrumental as separate stems for DAW workflows. Target explicit BPM values. Set song key. Blend multiple style references. This is production-grade control on top of a generative model — the closest thing to a DAW's mixing bus that a text-prompt system has shipped.

Honest vocal-quality caveat

Suno and Udio produce more polished, radio-ready vocals on average. Sonauto's vocals can sound stylized or occasionally unnatural on certain genres — the most common complaint in independent reviews. Trade-off: you accept sometimes-less-polished vocals in exchange for unlimited free generation and full commercial rights. For instrumental tracks the gap is much smaller.

Real position: if vocal polish is the axis you care about most, Suno wins. If free unlimited + full commercial + inpainting + stems + control are what you care about, Sonauto/Melodia is unmatched. For pure instrumentals, the vocal-quality gap disappears entirely.
Melodia · diffusion steps
CONTINUOUS SPACE
t=0
Noise
t=25
Denoising
t=50
Structure
t=75
Refining
t=100
Signal
→→→→
Melodia — latent-diffusion audio model · long-form music generation research paper (Housen & Tremblay)
v3
Who it fits

For creators who need commercial rights, not just demos.

The differentiator that matters most is not vocal quality — it's the licensing question. Sonauto is the only major free tier that lets you ship the songs you make.

Content creators

YouTube, TikTok, podcasts.

Background tracks, intros, outros, video game clip music. Generate as many drafts as you need, pick the take you like, ship. No monthly credit budget. No "wait until tomorrow" caps. No licensing paperwork.

Removes: royalty-free music library subscriptions.
Indie game devs

Ship the soundtrack.

User testimonial in the wild: "Generated my entire game soundtrack with Sonauto's free tier. Full commercial rights meant I could ship without licensing worry. Saved at least $2,000 in composer fees." That's the pitch.

Removes: composer contracts for small-budget titles.
Marketers

Jingles, ads, brand sonics.

A/B test campaign audio without paying per-license fees. Generate ten versions of an ad jingle, pick the winner, extend it, export stems for the audio engineer. Commercial use covered at $0.

Removes: stock-music licensing spend.
Producers

Stems + DAW pipeline.

Generate a starting point in Sonauto, split into stems, drop into Ableton, build a completely new arrangement around the AI vocal. It is a full remix pipeline in one tool — the producer surface most AI music tools skip.

Replaces: the "regenerate everything" workflow.
Everything in the box

Unlimited, commercial, controllable.

— Unlimited free

No daily cap. Ever.

Suno's free tier stops you at 50 credits per day. Udio caps too. Sonauto's free tier has no daily generation limit — you can iterate for hours until the take is right. This is the single biggest reason creators switch.

∞generations
— Full commercial rights

You own what you make.

Terms Section 8 (Output) assigns rights to you. Publish, stream, sell, license — Sonauto stays out of it. Every tier. Even free. This is what "unlocks commercial" means on Suno and Udio; here it's the default.

— Inpainting

Rewrite just the chorus.

Highlight a section of your generated song. Regenerate only that region. The rest of the track stays byte-for-byte the same. Every other major AI music tool forces full-song regeneration. This is the killer edit feature.

— Stem export

4 tracks. Separated.

Vocals Drums Bass Instr
— BPM + key

Target 128 bpm.

Explicit tempo control. Key selection. Structure editing. Multi-reference style blending. Producer-grade control on top of a generative model.

— MIDI conditioning

Upload a MIDI file.

Guide the model with your own chord progression, melodic contour, or rhythmic pattern. Melodia conditions its output on your input.

— REST API

Build on Melodia.

POST /v1/generations with prompt, tags, lyrics, and generation parameters. Bearer-token auth. Pay per generation. Ship your own product on top of the model.

# POST /v1/generations
"prompt": "lo-fi hip hop, 85bpm",
"prompt_strength": 0.85,
"instrumental": true,
"num_songs": 2
— Song extend

Add 60-80s or more.

Generated a 60-second clip you love? Extend it into a full-length track. Melodia continues the same musical structure and vibe rather than starting fresh. Perfect for turning a hook into a shippable song.

Honest comparison

Suno wins on vocal polish. Sonauto wins on everything else free.

Five AI music generators. Here's where each beats Sonauto — and where Sonauto wins back.

Capability Sonauto Suno Udio Riffusion MusicGen
Free tier generation cap Unlimited 50 credits/day 10 tracks/day Limited free Open-source, self-host
Full commercial rights on free tier Yes · included Paid tier only Paid tier only Depends on tier Model license
Model architecture Latent diffusion LLM-based LLM-based Spectrogram diffusion Autoregressive
Vocal polish quality Good · variable Best in class Excellent Weak vocals No vocals
Inpainting (partial regen) Yes · killer feature Not offered Not offered Not offered Not offered
Stem separation and export 4 stems export Premium Premium No No
Explicit BPM + key control Yes · v3 Text-hint only Text-hint only Limited Limited
MIDI / audio conditioning Yes Audio upload Audio upload No Yes · self-host
Public API REST · $0.05/gen Limited beta No public API Yes Self-host
Best for Producers · commercial · unlimited Polished vocals High-fidelity songs Experimental audio Research · self-host
Reviews · including the vocal-quality complaint

What producers are actually saying.

★★★★★

"The inpainting feature is what sold me. Every other AI music tool makes you regenerate the whole song if you don't like one section. Sonauto lets me rewrite just the chorus while keeping the verse I already love."

Independent producer · 5★ · switched from Suno for editing
★★★★★

"Sound quality is good, interface is simple, and unlimited free is unbeatable. But the AI-generated vocals can sound unnatural at times, and there's limited control over specific musical elements compared to more advanced software. It can also feel generic on certain genres."

Honest critical review · vocal quality trade-off · 4★
★★★★★

"I generated my entire game soundtrack with Sonauto's free tier. Full commercial rights mean I could ship the game without worrying about music licensing. That saved me at least $2,000 in composer fees."

Indie game developer · shipped title on free tier
— The disclosure

Where Sonauto actually sits.

This page is independent — built by Sonautoai.ORG as a reference and review site. Not affiliated with Sonauto, Inc. or Treblo. We tested the app on real prompts before writing this.

Sonauto was founded in 2023 by Ryan Tremblay (Cornell CS + History) and Hayden Housen (ML expertise), backed by Y Combinator. The technical bet was to use latent diffusion rather than the LLM-based approach dominant in AI music at the time. Their argument: continuous latent space gives users deeper control over musical elements than discrete audio tokens ever could. They published a research paper — "Long-form music generation with latent diffusion" — documenting coherent tracks up to 4:45. The model is called Melodia.

What's genuinely strong: unlimited free generation with no daily caps (Suno stops at 50/day); full commercial rights on the free tier, documented in Terms Section 8 (Output); inpainting for partial regeneration — the single feature producers cite most often for switching; stem separation and export for DAW workflows; explicit BPM and key targeting; MIDI and audio conditioning to guide the model with your own musical input; a public REST API at ~$0.05 per generation for developers; and a research-published architecture that gives the model a technical story other music generators can't tell.

Where it's honestly weaker: vocal quality is not at Suno or Udio's polished level — the most consistent criticism in independent reviews is that AI-generated vocals can sound stylized or unnatural on certain genres. On instrumental tracks the gap almost disappears. Advanced editing depth is limited compared to a full DAW — Sonauto is a generator plus a few production tools, not a replacement for Ableton or Logic. Learning curve for beginners is real despite "easy" marketing. Generation time varies — usually under 90 seconds full song, sometimes longer. And the model version toggle (v2.2 stable vs v3 preview) tells you the model is still maturing.

For the current product visit sonauto.ai, which redirects to Treblo. We are not affiliated with the publisher.

The Treblo rebrand disclosure.

In 2026 the team rebranded from Sonauto to Treblo. Their stated reason: "the new name is easier to say, spell, and remember." Same team, same Melodia model, same accounts. Existing Sonauto accounts continue to work. Some legacy branding lives on in the ecosystem — this reference site uses "Sonauto" since that is the historical, more-searched name. The product and the ownership terms are unchanged post-rebrand.

2023
FoundingRyan Tremblay + Hayden Housen ship Sonauto out of Y Combinator with the Melodia latent-diffusion model — bet on continuous-space audio generation vs LLM tokens
2024
Research paper"Long-form music generation with latent diffusion" documents coherent tracks up to 4:45 — technical differentiator vs closed-source competitors
2025
v3 + inpaintingMelodia v3 adds explicit BPM targeting, structure editing, multi-style-reference blending; inpainting ships as the killer edit feature
2026
Treblo rebrandRebranded to Treblo for a simpler name; native iOS + Android apps ship alongside the web + PWA surface; API opens to developers
Frequently asked

The short, honest answers.

Sonauto is an AI music generator that turns text prompts into complete songs with vocals and instrumentals. It is built on a proprietary latent-diffusion model called Melodia, developed by Ryan Tremblay and Hayden Housen — Cornell alumni backed by Y Combinator. The key differentiators are unlimited free generation (no daily caps) and full commercial rights included on every tier, even the free one. The service was rebranded to Treblo in 2026, but it is the same team, same songs, same accounts.
Yes, unlimited generation on the free tier — no daily cap, no watermark forced onto the audio, and full commercial rights included. The optional Pro tier speeds up the generation queue and unlocks higher-fidelity export formats, but Pro does not unlock rights that free users do not already have. This is different from Suno and Udio, which cap free tier generation and reserve commercial licensing for paid plans. The catch, if there is one, is that vocal quality on Sonauto is not always at Suno's polished level — see the honest review section on this page for detail.
Melodia is Sonauto's proprietary audio-generation model. It uses latent diffusion — the same class of technology behind image generators like Midjourney — but applied to audio. Unlike LLM-based competitors that predict discrete audio tokens, diffusion models operate in continuous latent space, which the Sonauto founders argue gives users more fluid control over musical elements. Their research paper, Long-form music generation with latent diffusion, documents generating coherent tracks up to 4 minutes 45 seconds. Version 3 of Melodia added explicit BPM targeting, structure editing, and multi-style-reference blending.
Inpainting lets you replace a specific section of a generated song without regenerating the whole track. If your chorus is perfect but the second verse is off, you highlight just the verse and regenerate that region — the rest of the song stays exactly as it was. This is the feature producers cite most in reviews. Every other major AI music tool forces you to regenerate the entire song if any section is wrong, which means losing the parts you liked. Inpainting is closer to a real production workflow.
Honestly, Suno and Udio produce more polished, radio-ready vocal takes. Sonauto's vocals can sound stylized or unnatural on certain genres — this is the most common complaint in independent reviews. The trade-off you accept for Sonauto's free unlimited generation and full commercial rights is that you sometimes need to run a few extra generations to get a vocal take you like, or use inpainting to fix a specific section. For instrumental-only tracks the gap is much smaller.
Yes. Sonauto's terms — Section 8 (Output) — assign ownership of generated content to you, the creator. You can publish, stream, sell, and license the songs you generate. Commercial rights are included on every tier, including the free tier. This is documented in the Terms of Service and is the same policy on Treblo (the post-rebrand entity). Users have generated full game soundtracks, YouTube video music, and podcast intros on the free tier and shipped them commercially.
In 2026 the team rebranded from Sonauto to Treblo. Their stated reason: the new name is easier to say, spell, and remember. It is the same team, same underlying Melodia model, and the same account — existing Sonauto accounts continue to work under Treblo. Some legacy references still exist across the web (this page uses Sonauto since that is the historical name). Both sonauto.ai and the new Treblo domain point to the same product.
Originally Sonauto was a web app with an iOS PWA installable via Safari (Add to Home Screen). After the Treblo rebrand, native iOS and Android apps were released. The web version remains the most feature-complete surface — inpainting, stem export, and API access are all web-first. The mobile apps are optimized for generation and playback rather than deep editing.
Yes. Sonauto ships a REST API. Create an account, generate an API key from the dashboard, and authorize requests with Bearer your_api_key. Submit POST requests to /v1/generations in JSON with parameters including prompt, tags, lyrics, prompt_strength, balance_strength, seed, instrumental (boolean), and num_songs. The API is priced per generation, roughly $0.05 per song, and is built for batch workflows, integrations, and building your own product surface on top of Melodia.
S

Type a prompt. Get a song. Ship it.

Free unlimited generation. Full commercial rights included. Latent-diffusion Melodia model. Inpainting, stem export, BPM control, MIDI conditioning, and a REST API for developers.

Try Sonauto → Get API access
Sonauto is a web-first AI music generator built on the Melodia latent-diffusion model by Sonauto, Inc. (now operating as Treblo). Free tier includes unlimited generation and full commercial rights. Pro tier offers queue priority and higher-fidelity exports at approximately $10/mo; API access is priced per generation. Regional pricing and features may vary. This page is independent and not affiliated with the publisher. Vocal quality on generated tracks varies — see honest review section above.