Ruach Studio: the YuE2 Studio for songs on your own GPU
Some of you already found our First Wave merge here: UCHIJ repacked RS-FW 4.0×0.7 for ComfyUI, and people have been downloading it. Thank you, UCHIJ, and thank you to everyone who tried it. This article is the story behind that file, from the people who made it.
What Ruach Studio is
Ruach Studio is an open-source (AGPL) studio for making whole songs on your own NVIDIA card. You write the style and the lyrics; the Studio composes, sings and renders the song with YuE2, the open song model from m-a-p, running through yue2.cpp. Nothing goes to a cloud and there are no credits: the model, the score and every take stay yours.
The rooms
Creator: the song itself, with a score you can edit before the audio
Writer: lyrics with a local or remote LLM
Refiner: 30+ nodes, stems, remaster, a piano roll and a DAW
Artist: cover art
Videographer: a clip for your track
Trainer: a LoRA on your own songs
Librarian: every take, with the settings that made it
Our finetunes
FW and TW are our two finetunes of YuE2 3B: the First Wave and the Tsunami Wave. The Tsunami Wave was trained on 109,250 tracks, about 6,650 hours, and the Studio's default model is RS-TW 2.0 ×0.75. The merges are published on Hugging Face under CC BY-NC-SA 4.0.
Running it
Today the Studio runs on Linux x86-64 with an NVIDIA card of 12 GB or more: one click in Pinokio, a single AppImage file, or from the source. A Windows package through WSL2 is being tested right now. macOS is not supported.
In ComfyUI, UCHIJ's repack of the First Wave loads in the standard YuE2 workflow.
Links
Site: https://ruachstudio.art
Studio models: https://huggingface.co/goldhub/Ruach_Studio_Models_v2
Questions, bugs and ideas are welcome in the comments or in GitHub issues.