What this is

We measured 6 minutes of song output generated in 126 seconds on an RTX 3090—Ruach Studio's real-world numbers. They also mark the first time open-source AI songwriting has run the complete "score-to-DAW (digital audio workstation)" chain, though ease-of-use still trails SUNO by a meaningful margin.

Ruach Studio is a local AI songwriting tool built around the open-source song model YuE2, released by the m-a-p team. In short: users write lyrics and style prompts, and the model composes, sings, and mixes on the user's GPU, producing full songs up to 8 minutes long.

What sets it apart is the workflow. The model first lays down the melody and chords in ABC notation (a text-based score format), which users can edit and re-render. LoRA training (a lightweight fine-tuning method) happens directly inside the studio. Post-production—stem separation, mastering repair, upscaling, lyric proofreading—runs as a single pipeline, finally exporting project files for professional DAWs like REAPER.

The code is open-sourced under AGPL-3.0; the model weights use CC BY-NC 4.0, meaning they cannot be sold commercially.

Industry view

Supporters see this as open-source AI music reaching "studio grade" for the first time—composition, scoring, mixing, and export form a closed loop that could in theory replace part of SUNO's workflow. The score-first design also answers musicians' longstanding criticism that AI "only outputs canned music."

The criticism we hear is equally pointed. The authors themselves admit the mixes hum, lyrics wander, and some instruments sound thin. To patch these issues, users have to train their own LoRAs—which pushes the bar up to real musicians, not casual hobbyists. Even more concretely, 24GB VRAM + Linux + 120GB storage is not a common configuration in Chinese households or small businesses, meaning in the short term the tool mainly serves two groups: hardcore AI music tinkerers, and content studios with strong on-premises deployment needs.

Another overlooked point: CC BY-NC 4.0 restricts commercial use, so songs produced with this tool cannot be sold directly. The path to commercialization is unclear.

Impact on regular people

  • For enterprise IT: No need to pay attention in the short term—this is a creative tool, not an office system.
  • For working professionals: Content operators, short-video creators, and self-media practitioners can stay on the sidelines—wait until a cloud-based substitute appears or a domestic vendor ships a simplified version before getting hands-on.
  • For the consumer market: It does not change the listening experience, but it confirms a trend: the cost of AI-generated content continues to fall, and the line between professional and amateur is being further compressed at the tooling layer.