Show HN: I built an AI music generator with some harnesses

Developer and music producer ShawWang launched onlymusic.ai, an AI music generation tool built with prompt preprocessing, a rule system, and model orchestration designed to make music creation accessible to people without a musical background, according to a Show HN post published on August 25, 2026.

The tool addresses three problems ShawWang observed among non-musician users, per the HN post: difficulty describing a desired sound, the habit of referencing an artist name rather than musical characteristics, and confusion caused by professional terminology in existing AI music interfaces. The solution is a layer of preprocessing that transforms simple user inputs into prompts the underlying model can use more effectively — what ShawWang calls "harnesses" around the base models.

onlymusic.ai offers two key preprocessing features: style translation and lyrics optimization. When a user names an artist, the system translates that into concrete musical descriptors — era, instrumentation, rhythm, vocal texture, and production style. For lyrics, it rewrites sections that may not generate well while preserving original rhyme, imagery, and intent. The UI itself has also been stripped of professional jargon, keeping only elements ShawWang judged necessary for a general audience.

The platform runs on open-source models, including ACE, with some models self-hosted to control costs. That infrastructure allows a free tier requiring no account: up to 5 songs in Fast mode before sign-in, and up to 20 per day after signing in. More capable models carry a lower daily generation quota. ShawWang noted in the HN post that the free quota may be reduced if infrastructure costs grow beyond current projections.

ShawWang states that several features remain under development, and that a non-form-based lyrics generator is currently in progress with a planned share to the HN community once complete. No funding, team size, or incorporation details were mentioned in the source.