Reverse Lo-fi Anime Girl — A Mini-Game Powered by Real-Time AI Music Generation and Music Sentiment Analysis

We’ve released Reverse Lo-fi Anime Girl, a playful mini-game built with real-time AI music generation and music sentiment estimation.

https://reverse-lofi.qosmo.jp/

In the game, players must keep generating music that helps the on-screen anime character stay focused on studying. To do so, you continuously catch falling text prompts and feed them into the AI music generator.

Game screen

If the music becomes too relaxed, the character falls asleep. If it gets too danceable, she stops studying and starts dancing. The key is to maintain just the right mood and energy by choosing the most appropriate text prompts. This game humorously inverts the famous lo-fi hip-hop streaming channels on YouTube that became an internet meme—channels designed to help human listeners concentrate while studying. Here, instead, humans become the players, working to keep the on-screen character focused by actively shaping the music. While you may think you’re “just playing,” you’re actually supplying valuable data that helps train and refine the AI. Gentle, study-friendly or sleep-friendly music is increasingly being replaced by AI-generated tracks. This work lightly visualizes that contemporary reality in a playful, thought-provoking way.

The character’s state changes depending on the music

Technical Background

The project combines Google Magenta Realtime API for generating music from text prompts in real time with a machine-learning model that estimates emotion in music. During a recent visit to Japan, a researcher from the Google Magenta team even mentioned in a talk that this was one of their favorite use cases of Magenta Realtime. It’s a great example of what becomes possible with real-time AI music generation. We hope you’ll give it a try.

Credits Concept/Prototyping: Nao Tokui Technical Direction: Ryosuke Nakajima Web Development: Chisato Takemura, Keito Takaishi Design: Mike Sound Design: Keigo Yoshida