Evaluation results on WildSongBench and SHS100K are published in the base model’s model card. Below are the evaluation results for full song generation on WildSongBench.<\/p>\n
\n
\u2192 Scroll horizontally to see all columns<\/p>\n
\n\n\n| Model<\/th>\n | Musicality \u2191<\/th>\n | SongBench Avg \u2191<\/th>\n | MuLan \u2191<\/th>\n | AllMusicCaps \u2191<\/th>\n | Q3O \u2191<\/th>\n | PER \u2193<\/th>\n<\/tr>\n<\/thead>\n |
\n\n| YuE 1<\/td>\n | 4.0847<\/td>\n | 4.9165<\/td>\n | 0.2623<\/td>\n | 0.2882<\/td>\n | 3.7301<\/td>\n | 36.38%<\/td>\n<\/tr>\n |
\n| SongBloom<\/td>\n | 3.4493<\/td>\n | 4.2350<\/td>\n | 0.2697<\/td>\n | 0.1926<\/td>\n | 3.0287<\/td>\n | 19.19%<\/td>\n<\/tr>\n |
\n| LeVo 2<\/td>\n | 5.4590<\/td>\n | 6.3247<\/td>\n | 0.3542<\/td>\n | 0.2680<\/td>\n | 3.9458<\/td>\n | 26.12%<\/td>\n<\/tr>\n |
\n| ACE-Step 1.5<\/td>\n | 5.1588<\/td>\n | 6.0118<\/td>\n | 0.4372<\/td>\n | 0.3869<\/td>\n | 4.5809<\/td>\n | 7.46%<\/td>\n<\/tr>\n |
\n| HeartMuLa<\/td>\n | 5.4963<\/td>\n | 6.2483<\/td>\n | 0.3823<\/td>\n | 0.2786<\/td>\n | 3.4907<\/td>\n | 10.71%<\/td>\n<\/tr>\n |
\n| DiffRhythm 2<\/td>\n | 4.4775<\/td>\n | 5.2428<\/td>\n | 0.3782<\/td>\n | 0.3255<\/td>\n | 4.0870<\/td>\n | 18.41%<\/td>\n<\/tr>\n |
\n| Muse<\/td>\n | 5.1692<\/td>\n | 6.0349<\/td>\n | 0.3937<\/td>\n | 0.3466<\/td>\n | 4.4038<\/td>\n | 33.42%<\/td>\n<\/tr>\n |
\n| MiniMax Music 3<\/td>\n | 5.3482<\/td>\n | 6.2830<\/td>\n | 0.3928<\/td>\n | 0.3609<\/td>\n | 4.4362<\/td>\n | 6.27%<\/strong><\/td>\n<\/tr>\n\nYuE2<\/strong><\/td>\n| 5.9075<\/td>\n | 6.7316<\/td>\n | 0.5068<\/strong><\/td>\n0.4054<\/strong><\/td>\n| 4.6819<\/td>\n | 8.44%<\/td>\n<\/tr>\n | \nYuE2 (best-of-8)<\/strong><\/td>\n6.2666<\/strong><\/td>\n6.9632<\/strong><\/td>\n| 0.5051<\/td>\n | 0.3980<\/td>\n | 4.7009<\/strong><\/td>\n| 9.79%<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/div>\n The base model YuE2 records the highest SongBench average among the evaluated open and proprietary models. On the other hand, other models excel in certain metrics, demonstrating a balance between generation quality and prompt compliance.<\/p>\n Strengths and Use Cases<\/h2>\nIn addition to generating music from lyrics and style prompts, the base model supports melody and chord planning, score editing via ABC notation, and agent-based song editing and revision. It supports Chinese and English lyrics, and is intended for zero-shot cover song generation and text-based music production.<\/p>\n Differences from Similar Models<\/h2>\nWhile Music Generation Model “YuE2-3B” Released, Producing Songs from Lyrics and Style<\/a> covered the standalone release of the base model “YuE2-3B”, Comfy-Org\/YuE2<\/code> covered in this article repackages those model files to fit the directory structure for ComfyUI.<\/p>\n<\/p>\n Hardware Requirements<\/h2>\nEstimated requirements (calculated by Local Model Watch)<\/strong> \u2014 3.6B parameters (taken from the base model m-a-p\/YuE2-3B)<\/p>\n | | | | | | | | |