AI Generated Music is HEATING UP!

Refusion’s open beta is positioning the company as a fast, beginner-friendly rival in the AI music race—while its “personalized modes” push toward a more tailored, user-specific music experience. The pitch is simple: unlimited free generations right now, a clean interface, and models that update quickly, with the creator’s testing centered on the jump from fuzz 0.6a (earlier access) to fuzz 0.8 (public beta). In a market where many competitors cap free output, that unlimited access lowers the barrier for experimentation and could accelerate adoption.

On the product side, Refusion looks familiar—prompt box, library, playlists, favorites, and social-style community feeds showing what others are generating. The standout difference comes in personalization. Refusion claims it learns a user’s “unique aesthetic” over time by tracking actions like creating songs, maintaining a streak, sharing invites, publishing, and favoriting enough tracks to unlock a “debut badge” that enables personalized modes. The creator acknowledges the incentive structure—more platform use means more data—but also treats the feature as genuinely valuable because it aims to steer future generations toward a user’s tastes. The beta also includes an open approach to lyrical content, including swearing in prompts, contrasting with the more restrictive safety behavior common in some LLM-driven systems.

When it comes to generation, the workflow is streamlined. A basic prompt can produce full-length songs (about three minutes) quickly, with lyrics and song structure handled automatically. For more control, Refusion offers a compositional mode for adding lyrics (or using its “Ghostwriter”) and sound/vibe prompting, plus an instrumental mode for background tracks. The creator repeatedly notes that results can vary—AI music still involves randomness—but highlights that Refusion’s structure changes feel more dynamic than some competitors, with clearer shifts between sections like verses and choruses.

Refusion’s lyric behavior shows both strengths and quirks. The Ghostwriter avoids directly referencing copyrighted characters or brands in a test involving “Grimace” and “McDonald’s,” generating a workaround that the creator found less satisfying. In other experiments, the model inserted an unexpected Spanish verse without explicit prompting, illustrating how creative drift can produce surprises. The creator also used external lyric generation (ChatGPT) and then tuned sliders like “lyrics strength,” plus “weirdness,” and multi-part sound timing (e.g., specifying an intro that starts at 0 seconds and ends at 17 seconds). That timing control—separating intro/verse/chorus-like elements—was treated as a notable capability.

Uploads add another layer: Refusion can extend or remix using user-provided audio, with options to extend, cover, or replace parts. In practice, the creator found remixing wasn’t always reliable—sometimes the system pulled in the wrong style or even old prompt content—leading to a refresh and a workaround. Still, the overall impression is that Refusion is competitive on clarity (voices and lyric delivery) and is actively iterating.

Overall, the creator frames Refusion as “close” to Udio and Sunno in quality, with some areas where it may lag musically, but with enough differentiators—free unlimited beta, rapid model updates, section switching, and personalization—that it could pressure incumbents. Looking ahead, the creator argues 2025 could be a breakout year for AI music, especially as other major players like ElevenLabs prepare music-related releases.