☕ One coffee = one happy creator: https://buymeacoffee.com/the_frizzy1
YuE2 is a 3B open music model from the Multimodal Art Projection team. You give it a music style and your lyrics, it gives you a complete song with vocals and instruments. Cover mode puts a new style and new lyrics on your own song's melody, from the same node.
What you need
✔ Custom nodes — None. Core ComfyUI only.
✔ ComfyUI version — 0.39 or newer (YuE2 support)
✔ VRAM — 7.5 GB peak — tested, not estimated
✔ System RAM — 4.0 GB peak
✔ Tested on — RTX 3060 12 GB · 32 GB DDR4 · ComfyUI 0.39.2 (Comfy Desktop)
✔ Licence — Workflow MIT · YuE2 and SheetSage2: CC BY-NC 4.0 (non-commercial)
This is the lightest thing on my channel. 7.5 GB peak means an 8 GB card should handle it.
Models to download
Click a file name to download it. Put it in the folder shown, inside ComfyUI/models/.
Required — Comfy-Org/YuE2. SheetSage2 is only for cover mode.
checkpoints/
File: yue2_3b_int8_convrot.safetensors — 4.0 GB (all modes)
Download:
https://huggingface.co/Comfy-Org/YuE2/resolve/main/checkpoints/yue2_3b_int8_convrot.safetensors
Put it here:
ComfyUI/models/checkpoints/
audio_encoders/
File: sheetsage2_bf16.safetensors — 1.4 GB (Cover mode)
Download:
https://huggingface.co/Comfy-Org/YuE2/resolve/main/audio_encoders/sheetsage2_bf16.safetensors
Put it here:
ComfyUI/models/audio_encoders/
The bf16 checkpoint (yue2_3b_bf16, 7.8 GB) works too — pick it in the Model field.
Modes
Text to music takes a music style plus lyrics. With ABC planning on, YuE2 first writes the melody and chords as an ABC score (median 13 s), then makes the song from it. Off goes straight to the song.
Cover (your song) takes your song plus a new style and new lyrics: SheetSage2 transcribes your melody to ABC (8 s), YuE2 sings the new lyrics on it.
The score that was used is shown in ABC score used — copy it, edit it, reuse it. Lazy switches pick the path, so SheetSage2 and the ABC planner only load when the mode uses them.
Speed on my RTX 3060
Same style, lyrics and seed as the official templates. Times are queue to finished; the first run includes loading the model.
Text to music, ABC planning on — 1:47 song length, 58 s rtx 3060
Text to music, ABC planning off — 1:38 song length, 40 s rtx 3060
Cover, the template's reference song — 1:40 song length, 51 s rtx 3060
Medians: writing the music 24 s, sampling 17 s.
What caught me out
Max length (s) is only a cap. The length comes from your lyrics — 120 s and 240 s gave me the same song.
Put
[Verse],[Chorus],[Bridge]above the sections, or you don't get the structure."Invalid audio file" before it starts: Your song is un-bypassed but its file is not on your PC. Load it, or bypass it (Ctrl+B) for text to music.
"Cover needs your song": you picked Cover and Your song is still bypassed.
The licence bites on this one. YuE2 and SheetSage2 are CC BY-NC 4.0, so non-commercial.
Get the workflow
GitHub: https://github.com/Thefrizzy1/The_frizzy1-Workflows/tree/main/workflows/audio/yue2-music-12gb
Hugging Face mirror: https://huggingface.co/The-frizzy1/The_frizzy1-Workflows/tree/main/workflows/audio/yue2-music-12gb
More videos on this stuff: https://www.youtube.com/playlist?list=PLcgjaTzmjLkE