SELECTED PROJECTGenerative AI

Procedural osu! Beatmap Generation with Rectified Flow

A Rectified Flow diffusion model for procedural osu! beatmap generation and informal comparative evaluation.

62%INFORMAL PREFERENCE
  • Python
  • PyTorch
  • Rectified Flow

Objective

Explore diffusion-based generation as a way to produce playable osu! beatmaps procedurally.

System

Trained a diffusion model with Rectified Flow using Python and PyTorch, then compared its generated maps with outputs from the osu-dreamer project.

Outcome

Across 52 informal A/B comparisons, maps from this model were preferred in 62% of comparisons. This result is exploratory user feedback rather than a formal user study.