TuneTree

Inspiration

Current AI music generators make it easy to generate something, but hard to create your song. We noticed that current tools produce generic lyrics and melodies, have cluttered interfaces, and force users to regenerate entire songs just to tweak small details. Editing notes, lyrics, or styles often requires starting over, which breaks creative flow. TuneTree was inspired by the idea that music creation should feel iterative, personal, and controllable, more like growing a song than generating a one-off output.

What it does

TuneTree is an AI-powered music creation platform that gives users intuitive, granular control over how their music evolves.

Users begin by describing their idea, selecting genres, and choosing whether to include lyrics. TuneTree then generates two song variations side by side, displaying both the lyrics and stylistic elements of each. Instead of repeatedly regenerating entire tracks, users can directly edit lyrics and styles, then choose the version they prefer.

The system regenerates the unselected track to be more similar to the chosen one and less similar to the rejected one. As users continue making pairwise selections, the music progressively aligns with their preferences while preserving the best version across generations.

When satisfied, users can move into Studio Mode, where they can download tracks as MP3, convert audio to MIDI, and either export the files or open them in an in-browser editor to modify the actual notes without regenerating the song. Users can also publish tracks to their portfolio.

The Portfolio page lets users play, manage, and delete past creations, making it easy to track progress and showcase finished work. Overall, TuneTree enables users to iteratively refine music through preference-based feedback, direct editing, and flexible export options, turning AI-generated ideas into fully customizable, portfolio-ready tracks.

How we built it

We built TuneTree with a modern, scalable stack. The frontend is built with Next.js and React for a clean, responsive user experience. The backend uses FastAPI and Supabase for data storage, authentication, and file management. For AI-powered music generation, we integrated ElevenLabs for audio synthesis, OpenAI for structured lyric and composition planning, and Basic Pitch for audio-to-MIDI conversion. This architecture allowed us to separate creative logic, generation, and editing into modular steps that support iteration.

Challenges we ran into

One major challenge was designing a system that supports iterative creativity rather than one-shot generation. Coordinating multiple AI services while keeping metadata consistent across versions required careful orchestration. We also had to balance flexibility with usability, giving users deep control without overwhelming them. Handling audio formats, MIDI conversion, and editable note representations introduced additional technical complexity, especially under hackathon time constraints.

Accomplishments that we’re proud of

We’re proud that TuneTree goes beyond simple music generation and enables real creative control. Our side-by-side version comparison, editable lyrics and styles, and integrated MIDI editor directly address the biggest pain points we identified in existing tools. Building a full end-to-end pipeline, from prompt to audio, to MIDI, to downloadable and editable output, within the scope of Devfest was a major accomplishment.

What we learned

Through building TuneTree, we learned how important structure and feedback loops are when working with generative AI. Small UX decisions can dramatically affect how “creative” a tool feels. We also gained hands-on experience integrating multiple AI systems into a cohesive product and learned how to design interfaces that encourage exploration rather than one-off usage.

What’s next for TuneTree

Next, we plan to introduce an upgraded agent that maintains persistent preference memory across sessions, allowing TuneTree to continuously learn from a user’s past likes and dislikes rather than starting fresh each time. We could also model the agent after after tree-based decision agents, similar in spirit to systems like AIDE or RD-Agent, where each generation represents a node and user feedback guides structured exploration and refinement over time.

We also aim to support real-time audio streaming during creation, one-click global publishing to platforms like Spotify, Apple Music, and TikTok, and a public showcase where users can explore, share, and discover each other’s portfolios.

Ultimately, TuneTree will evolve into a Cursor-style copilot for digital audio workstations (DAWs), an adaptive, personalized system that helps users compose, edit, and refine music directly inside their creative workflow, rather than just generating tracks. It’s not just an AI music generator, but a fully integrated music creation ecosystem.