unknown

ACE-Step 1.5 is an open-source music generation model that runs locally with under 4GB VRAM, generating full songs in under 2 seconds on A100 or under 10 seconds on RTX 3090. The model uses a hybrid architecture combining a Language Model planner with a Diffusion Transformer, supporting features like multi-language lyrics (50+ languages), cover generation, audio editing, track separation, and LoRA fine-tuning from just a few songs. It offers flexible duration control (10 seconds to 10 minutes), batch generation, and various editing capabilities including vocal-to-BGM conversion and selective audio repainting. Available through Gradio UI, Python API, or REST API with automatic model downloads from HuggingFace or ModelScope.

ace-step/ACE-Step-1.5: The most powerful local music generation model that outperforms most commercial alternatives

Exploring advanced technology in platforms, DevOps, Kubernetes, IaC, open-source AI tools, and more. Innovation and collaboration.