Course 01
Agentic Music Generation
Teach an AI to compose. Then teach it taste.
Two weeks, about half an hour a day. You will build an agent that takes a one-line brief, drafts a track, listens back, critiques itself, and iterates until the music is good. No music theory required. Basic Python helps but is not required.
What you
walk away with.
Not a certificate. Working software, and the taste to judge what it makes.
A working agent that turns a text brief into a finished two-minute track
A prompt library for the genres and moods you actually care about
An evaluation loop that tells a good generation from a bad one without you listening to every take
A production pipeline with a known cost per track and rights you can publish under
Outline
Two weeks, one agent.
This is the public outline. Your path is generated around what you already know, so lessons you do not need get cut and the two weeks stay realistic.
Foundations and prompting
Understand what the model hears, and learn to steer it with words, references, and a scorecard.
- 01
How machines make music
Day 1-2You cannot steer a model you do not understand. This gives you the mental model behind every generation.
- What a music model actually hears · 14 min
- Autoregressive vs diffusion, in plain words · 15 min
- Why most generations sound generic · 12 min
- 02
Prompting for sound
Day 3-4Genre, tempo, instrumentation, structure, lyrics. Learn the words that move the model.
- Anatomy of a strong music prompt · 16 min
- Instrumentation and arrangement words · 14 min
- Writing lyrics the model can sing · 18 min
- 03
Conditioning and references
Day 5A hummed line or a reference clip gets you closer than any paragraph of adjectives.
- Melody and reference-audio conditioning · 16 min
- Style without copying · 12 min
- Your prompt library · 14 min
- 04
Judging a generation
Day 6-7An agent is only as good as its critic. Build the scorecard before you build the loop.
- Listening tests that catch the obvious · 12 min
- Audio embeddings as a critic · 18 min
- Your first automated scorecard · 16 min
Agents and shipping
Turn one-shot generation into a loop that plans, judges, iterates, and ships a finished track.
- 05
Your first composing agent
Day 8-9The leap from one-shot generation to an agent: plan, generate, judge, regenerate.
- The plan, generate, judge loop · 20 min
- Giving the LLM a music model as a tool · 18 min
- Stopping rules · 12 min
- 06
Arrangement and stems
Day 10-11Real tracks have sections, layers, and transitions. Teach the agent to think in parts.
- Stem separation and layering · 18 min
- Multi-section tracks that hold together · 20 min
- Transitions, intros, outros · 14 min
- 07
Iteration with feedback
Day 12Nobody ships take one. The agent has to take a note and act on it without starting over.
- Turning a human note into a model edit · 15 min
- Memory across takes · 14 min
- Regenerate only what is wrong · 16 min
- 08
Ship it
Day 13-14Cost, latency, rights. The difference between a demo and something people can use.
- A production pipeline under a dollar a track · 18 min
- Licensing, rights, and what you can publish · 14 min
- Your final brief-to-track build · 30 min
Who it is
built for.
Producers and musicians curious what AI can do beyond presets
Developers who want to build creative tools, not just chatbots
Founders prototyping a music product before hiring an audio team
Early access
Doors open 22 Nov, 22:22 PM IST.
Sign up now, pick the skills you care about, and your dashboard counts down to the moment the course goes live.