🎵 ACE-Step 1.5 Custom Edition
Three powerful interfaces for music generation and training
Models will download automatically on first use (~7GB from HuggingFace)
Full-featured standard ACE-Step 1.5 interface
Audio Input (Optional)
Upload audio file or record to use as style guidance
Generated Output
Advanced Controls
Custom Timeline-based Generation
Generate 32-second clips that seamlessly blend together on a master timeline.
How far back to reference for style guidance
Train Custom LoRA Models
Step-by-step wizard: provide audio data, auto-label with LLM, preprocess, and train.
Download audio from a HuggingFace dataset repo. Labels and progress are synced back to the repo automatically.
Auto-label samples using the LLM, then review and edit metadata.
Sample Review (editable: BPM, Key, Caption)
# | Filename | Duration | Lyrics | Labeled | BPM | Key | Caption |
|---|---|---|---|---|---|---|---|
Encode audio through VAE and text encoders to create training tensors.
Configure and run LoRA training.
Tips
- Upload 10+ audio samples for best results
- Keep samples consistent in style/quality
- Higher rank = more capacity but slower training
- Default settings (rank=8, lr=1e-4, 100 epochs) are a good starting point
About
ACE-Step 1.5 Custom Edition by Gamahea | Based on ACE-Step