ACE Step is an open-source AI music generation model co-developed by ACE Studio and StepFun. Released under the MIT license, it is the first open-source model to approach commercial platform quality, capable of generating complete songs with vocals from text prompts in under 2 seconds on professional hardware.
The latest version, V1.5, uses a hybrid Language Model + Diffusion Transformer architecture. The LM acts as a planner that transforms queries into song blueprints, while the DiT generates the actual audio. It supports durations from 10 seconds to 10 minutes, 50+ languages, and 1000+ instruments and styles.
ACE Step can run on consumer hardware with less than 4GB VRAM, supports LoRA fine-tuning from just 8 songs, and integrates with ComfyUI, VST3 plugins, and REST APIs. On MeloLab, you can access ACE Step alongside other leading models without any local setup.