The archive · AI & Models · Product decision · 2024–2026
Sonauto's API bet: open its music model to developers, keep building as Treblo
Sonauto launched free unlimited AI song generation in 2024, opened its full model to developers in 2025, and now sells Melodia v3 as Treblo.
Sonauto
What the business is
Generative-music startup: a free, unlimited text-to-song web app plus a paid API that gives developers access to its own music model.
How it started
Two co-founders launched Sonauto around May 2024 (they were ten months past launch in March 2025) with a text-to-song web app that was free and unlimited. In their launch post they framed the race as two people competing on "the best model and average user UI", and described users and pro musicians asking for DAW plugins, live-performance transition generators and other features the pair had no time to build.
What happened
On 2025-03-03 the founders posted "Show HN: Sonauto API" to Hacker News, opening full access to the model — generation, inpainting, extensions, transition generation, inverse sampling — with Python examples, while keeping consumer generations free. The thread drew 127 points and 157 comments; the founders answered pricing and model questions directly, as debate ran from hobbyist unlimited plans to whether training on artists' work is theft. Early test users had already built on the API: Lemon Slice used it for a singing-to-video product and a musician user published an open-source wrapper.
No ending yet — it is still running.
Background
Sonauto is a generative-music startup run by two co-founders. Around May 2024 they launched a web app that turns text prompts into songs, free and unlimited, and spent the next ten months making the underlying latent-diffusion model bigger and better — by March 2025, they said, it had a much stronger GAN decoder. Their bet was that in the music-AI race, model quality alone could carry a two-person company.
On 2025-03-03 they posted "Show HN: Sonauto API": full access to generation, inpainting, extensions, transition generation and inverse sampling, with Python examples and the consumer product still free and unlimited. The thread reached 127 points and 157 comments, and the founders answered questions on pricing while defending training on artists' work in a long copyright debate. Early users had already built integrations — Lemon Slice's singing-to-video tool and an open-source wrapper by a musician user.
By the 2026-09 capture the product was being sold under the Treblo brand: Melodia v3 with streaming output roughly 15 seconds to first audio, 4,160 style tags, iOS and Android apps, and API plans from $11/month after a 1,500-credit free trial. Treblo's own FAQ says the name changed because it is easier to say, spell and remember — same team, same songs, same account. Revenue is not public, so whether the API pays for the free consumer tier remains open.
What has to be true
- The API turns two founders' missing capacity into other people's products: DAW plugins, live-transition tools and singing-to-video were built by outsiders once the model was exposed.
- It monetizes without rationing the consumer product: free unlimited generation stayed the demo and funnel, while paid API credit plans became the revenue surface.
- HN reception was concrete and inspectable: 127 points and 157 comments on 2025-03-03, with founder replies and working third-party integrations in the thread.
- The story kept moving: by 2026 the same team had shipped a third-generation model under a new brand, showing the API launch was a platform bet, not a one-off.
What can be applied
When a tiny team cannot build the long tail of features, open the platform: outsiders ship those features as products, and API revenue funds the free tier that made the model known.
Aftermath
As of 2026-09-05 the venture is live as Treblo: Melodia v3 is sold through api.treblo.com with streaming output roughly 15 seconds to first audio, iOS and Android apps at treblo.com, 4,160 style tags, and developer plans from $11/month up to $330+ after a 1,500-credit free trial. The page at sonauto.ai/developers now describes a language-model architecture with dramatically lower compute costs — a shift from the diffusion design of 2025. No revenue or usage figures are published, so the API's ability to fund the free consumer tier remains unproven.
Sources
spotted an error? The archive wants to know.
Your turn
You just read one. Describe what you are building, and see who is betting on the same thing.
Free account · 3 free questions · no card