ACE-Step 1.5

ACE-Step 1.5 is an open-source music generation foundation model developed by StepFun & ACE Studio. It combines a Language Model (LM) planner with a Diffusion Transformer (DiT) to produce commercial-grade music.

Features

• Text-to-Music: Generate music from text descriptions (genre, mood, BPM, key, etc.)
• Lyrics-to-Song: Generate songs from structured lyrics with timestamps
• Audio Repainting: Edit specific sections of existing audio
• Audio Continuation: Extend existing audio clips
• LoRA Training: Fine-tune with your own audio data

Startup Mode

Set ACESTEP_MODE to choose the service to launch:

Mode	Port	Description
gradio (default)	7860	Interactive Gradio Web UI for music generation
api	8001	REST API server for programmatic access (/generate, /health)

GPU Requirements

VRAM	Mode	LM Model
6 GB	DiT-only (no LM)	None
6-8 GB	DiT + LM	acestep-5Hz-lm-0.6B
8-16 GB	DiT + LM	acestep-5Hz-lm-1.7B
16+ GB	DiT + LM	acestep-5Hz-lm-4B

Configuration

• ACESTEP_MODE: Startup mode — gradio (Web UI) or api (REST API server)
• ACESTEP_CONFIG_PATH: DiT model variant (default: acestep-v15-turbo)
• ACESTEP_LM_MODEL_PATH: LM model size (default: acestep-5Hz-lm-1.7B)
• ACESTEP_LLM_BACKEND: LLM backend — vllm (faster, needs 8GB+) or pt (PyTorch fallback)
• ACESTEP_API_KEY: Optional API authentication key

Links

• GitHub Repository
• Model Zoo
• API Documentation