yue vs ace-step-1.5
Two of the top projects in Open models, side by side: who leads on each score part and who each one suits, then setup, license, activity and what each review found.
yue
Lyrics-to-song model that writes an editable melody and chord score first
ace-step-1.5
Text-to-music model with LM planner, DiT decoder, Gradio UI and REST API
Verdict
yue ranks higher, 80 to 73. Of the 4 score parts we measure for both, yue leads on 1, ace-step-1.5 on 2 and 1 is even.
- Adoptionace-step-1.5 leads by 3 points, 79 to 76: close to even.
- FreshnessEven at 100.
- Maintenanceyue leads by 42 points, 91 to 49: a wide gap.
- Agent-readyace-step-1.5 leads by 45 points, 45 to 0: a wide gap.
Pick yue if you are
Researchers and musicians running local song generation on a 24 GB NVIDIA GPU
- Editable ABC score between prompt and audio lets you inspect and change melody and chords
- Staged Python API: plan, generate_semantic, synthesize, decode; artifacts saved per run
Pick ace-step-1.5 if you are
Musicians and developers generating music locally, with LoRA fine-tuning
- Runs locally with under 4GB VRAM using the DiT-only turbo model; README claims under 10s per song on RTX 3090
- Covers text-to-music, cover, repaint, stem separation, and vocal-to-BGM in one model family
| What we compare | yue | ace-step-1.5 |
|---|---|---|
| Score parts, out of 100 | ||
| Adoption | 76, popular | 79, popular |
| Freshness | 100, active | 100, active |
| Maintenance | 91, healthy | 49, patchy |
| Agent-ready | 0, none | 45, minimal |
| Facts from GitHub and the README | ||
| Stars | 13.6k | 15k |
| License | Apache-2.0 (permissive) | MIT (permissive) |
| Last commit | Nov 2026 | Aug 2026 |
| Last release | Sep 2026 | May 2026 |
| Language | Python | Python |
| Docker | No | Yes |
| GPU | Required | Optional |
| arm64 or Apple Silicon | Mentioned | Mentioned |
yue
YuE2 generates full songs with vocals and accompaniment from lyrics and a style prompt. It first plans an editable ABC score (melody and chords), then renders semantic tokens and acoustic latents to 48 kHz stereo audio with an AR–NAR Mixture-of-Transformers and a VAE. The repo also covers zero-shot covers via SheetSage2 transcription, score editing, and an agent skill. Code is Apache-2.0; model weights are CC BY-NC 4.0 with an additional creator permission.
Who it is for: Researchers and musicians running local song generation on a 24 GB NVIDIA GPU
Strengths
- Editable ABC score between prompt and audio lets you inspect and change melody and chords
- Staged Python API: plan, generate_semantic, synthesize, decode; artifacts saved per run
- Zero-shot covers via SheetSage2 transcription with the same checkpoint, no cover fine-tuning
- Reports WildSongBench results against Suno, Mureka and others, with a reproduction guide
Weaknesses
- Needs Linux, Python 3.12 and an NVIDIA BF16 GPU with 24 GB VRAM
- Weights are CC BY-NC 4.0; commercial company use requires a separate license
- Editing renders a new recording; the original waveform is not preserved
- Benchmark numbers are automatic metrics from the authors; gaps between top means are not significant
- GPU required
- Needs Hugging Face model downloads, SheetSage2 (separate environment, for covers), PyTorch
- Models: YuE2-3B, YuE2-Vae, YuE2-Vae-legacy, SheetSage2, MERT-v2
ace-step-1.5
ACE-Step 1.5 generates full songs from text prompts and lyrics using a Qwen3-based language model that plans the song and a diffusion transformer (2B or 4B XL) that renders audio. It also does cover generation, repainting, stem separation, vocal-to-accompaniment and LoRA training. It ships a Gradio UI on port 7860, a REST API on port 8001, a CLI and a Python API.
Who it is for: Musicians and developers generating music locally, with LoRA fine-tuning
Strengths
- Runs locally with under 4GB VRAM using the DiT-only turbo model; README claims under 10s per song on RTX 3090
- Covers text-to-music, cover, repaint, stem separation, and vocal-to-BGM in one model family
- LoRA training from a few songs: 8 songs in about 1 hour on a 12GB 3090, from the Gradio UI
- Several interfaces: Gradio UI, REST API, CLI, Python API, and a separate VST3 plugin
Weaknesses
- XL 4B models need at least 12GB VRAM with offload and quantization, 20GB for comfortable use
- Low-VRAM setups (6GB or less) run without the LM, losing planning and query rewriting
- Quality claims (between Suno v4.5 and v5) are the authors' own, with no numbers in the README
- Python 3.11-3.12 only; CUDA recommended, other backends (MPS, ROCm, XPU, CPU) are secondary
- GPU optional
- Docker + Compose
- Needs uv, Python 3.11-3.12, Gradio
- Models: ACE-Step DiT 2B (base, sft, turbo), ACE-Step XL DiT 4B (base, sft, turbo), acestep-5Hz-lm 0.6B/1.7B/4B (Qwen3-based), LoRA adapters
- port 7860
More in Open models
Building with these?
We help teams pick an open-source AI stack and run it in production. Client work never changes a rank.