LLM-written music. Try it live: audial.dev
Music has shaped humanity since our beginning, pushing us towards futures once thought impossible. Now it's time for AI to unlock what comes next.
And with all new things, there is much exciting work to be done here. Upwards and onwards!
# clone and install
git clone https://github.com/DorsaRoh/audial.git
cd audial
pnpm install # or: npm install
# start the dev server
pnpm dev # or: npm run devOpen http://localhost:3000, click ⚙️ (settings), choose your model and enter your api key, and begin!
It is rare for a model to produce a great song on the first prompt. In practice, you have to vibe with the model!
What works:
- iterate, don't restart - refine the musical idea
- react to what it generates (for example, if a part is good, try asking it to expand it)
- let the model settle into a direction instead of forcing novelty every turn
-
Unless you explicitly provide the musical material (notes, structure, midi, code, etc.), the model cannot reliably recreate specific existing works.
That means prompts like "write a theme song for Game of Thrones" will usually fail
If you want accuracy, you must provide concrete musical constraits or examples
There is much exciting work to be done. The future is bright!
known bugs: audio not working on mobile