01
Our own sound material
The pipeline runs on sound we own, not on whatever a public model absorbed. That is a licensing position as much as an aesthetic one: nothing in the catalogue depends on material we cannot account for.
An Aunimeda engineering project
A pipeline that ships music, not demos
Aunimeda is a software company. In 2022 we wrote our own pipeline for music generation, curation, mastering and distribution, and ran it on sound material we own — not prompts into someone else's service. It has since shipped 76 tracks across four platforms. Everything below opens and can be checked.
76tracks shipped
24releases
4distribution platforms
2022first release
Signal chain
Five stages, and we own all five
The “AI music” market is mostly a prompt box on someone else's server. The difference matters: you cannot put a quality gate, a retry policy or a release schedule inside a product you rent. No stage below is optional — platforms reject releases with out-of-spec loudness or broken metadata — and that is exactly why this pipeline works as an honest test of a system rather than a showcase.
02
A pipeline we wrote
Not a prompt box on someone else's server. Owning the pipeline is what makes the rest of this list possible at all: you cannot put a quality gate, a retry policy or a release schedule inside a product you merely rent.
03
Curation
The hard part is throwing away. Candidates are scored against the brief and against each other; only a small share of generated material survives to mastering. This is where a coherent series comes from instead of 76 unrelated experiments.
04
Mastering & metadata
Loudness normalisation to streaming targets, true-peak limiting, then ISRC codes, credits and genre taxonomy. Platforms reject or silently re-encode anything outside spec, and bad metadata is the most common reason a release fails ingestion.
05
Distribution
Delivery to Spotify, Apple Music, Amazon Music and YouTube, then reconciliation: every release verified as live on every platform. Four independent reviews we do not control — which is what makes this catalogue evidence rather than a portfolio claim.
Controllable output
57% of the catalogue is one coherent series
Generating a lot of tracks is easy. Holding a brief is not: 43 of 76 tracks sit in one niche — tabletop RPG scoring — and sound like one body of work rather than 24 independent attempts. That controllability is exactly what clients mean when they ask for generative AI that fits a spec.
- LPDragons in Dungeons
- LPBards in Dungeons
- LPDwarves in Dungeons
- LPTavern
Catalogue
24 releases, 7 of them albums
The full catalogue as listed on Apple Music. Every row links out: verifiable in one click.
SGSadness of the Universe
SGInktober
SGNight Charm
SGGrust
SGEchoes
SGFlow
SGWater Party
SGMiku's Childhood
SGMeow Meow Meow Meow Meow
SGTijuana
SGNeon Serenade
SGTelegram
SGPika Gal
SGNight Sky
SGStarLight
SGWarm Hot
LPDwarves in Dungeons
LPHorrorify
LPBards in Dungeons
LPCyberpunk
LPDragons in Dungeons
LPTavern
LPI
SGBeach Party
Distribution
Live on four independent platforms
Each runs its own ingestion review. Passing all four, twenty-four times over, is the kind of proof a slide deck cannot produce.
What this makes us good at
Generative AI in production, not in a demo
The same architecture — prompt design, model selection, automated quality gates, post-processing and delivery to third-party APIs — applies equally to text, image, audio and video workloads. Aunimeda has been building websites, mobile apps, chatbots and custom AI solutions since 2010, from Bishkek and Los Angeles.
FAQ
Common questions
Is Aunimeda a music label or a software company?
Aunimeda is a software company, founded in 2010, building websites, mobile apps, AI chatbots and CRM systems. Aunimeda Music is one of its internal engineering projects: a production pipeline for AI-generated music. The music is the output that proves the system works.
Why does an IT company run a music pipeline?
Because shipping to Spotify, Apple Music and Amazon Music is an unusually honest test of an AI system. It has external acceptance criteria we do not control: metadata validation, loudness specification, artwork requirements and ingestion review by four independent platforms. A demo can be staged; 24 releases live on four platforms cannot.
Do you just type prompts into an AI music service?
No. The pipeline is written in-house and runs on our own sound material. That distinction is the whole point: a hosted prompt box gives you output, but you cannot put quality gates, retry policies or a release schedule inside a product you rent — and you cannot account for the provenance of material you did not source. The brief, the curation and the release process are operated by people; the generation runs on infrastructure we own.
Can Aunimeda build a generative AI pipeline for my product?
Yes. The same architecture — prompt design, model selection, automated quality gates, post-processing and delivery to external APIs — applies to text, image, audio and video workloads. Aunimeda ships custom AI solutions, chatbots and integrations for clients from its Bishkek and Los Angeles offices.