AI TL;DR
AMD enables local AI music generation with ACE-Step 1.5 on Ryzen and Radeon. No cloud fees, royalty-free output, and commercial use allowed. Here's the complete guide.
Forget paying Suno or Udio per track. AMD just enabled local AI music generation on your own hardware.
The ACE-Step 1.5 model runs on AMD Ryzen AI processors and Radeon GPUs, generating full songs with vocals in as little as 30 seconds—completely offline and royalty-free.
What is ACE-Step 1.5?
ACE-Step 1.5 is an open-source AI music generation model released on February 3, 2026. It combines:
- A lightweight language model
- A diffusion transformer
- Vocal synthesis capabilities
The result? Full songs from prompts, lyrics, or reference audio—running entirely on your local machine.
Key Features
| Feature | Detail |
|---|---|
| VRAM Required | Less than 4GB |
| Song Length | Up to 10 minutes |
| Languages | 50+ for lyrics |
| License | MIT (commercial use OK) |
| Training Data | Royalty-free only |
| Output | No copyright claims |
Why This Matters
1. No Per-Track Fees
Cloud services like Suno and Udio charge for generation:
- Subscription fees
- Per-song limits
- Premium features locked
With ACE-Step locally: generate unlimited music for free.
2. No Internet Required
Perfect for:
- Remote work locations
- Privacy-sensitive projects
- Offline creative workflows
- Areas with limited connectivity
3. Commercial Use Allowed
The MIT license means:
- Use generated music in videos
- Sell your creations
- Include in commercial products
- No attribution required
4. No Copyright Issues
ACE-Step was trained exclusively on royalty-free, non-copyrighted material. This means:
- No surprise copyright claims
- Safe for YouTube/TikTok
- No DMCA takedowns
- Clean chain of title
Performance Benchmarks
How fast can you generate music?
| Hardware | 2-Minute Song Time |
|---|---|
| AMD Radeon AI PRO GPU | ~30 seconds |
| AMD Ryzen AI Max+ 395 | ~66 seconds |
| Consumer Radeon GPU | 1-2 minutes (varies) |
That's fast enough for real-time creative iteration.
Getting Started
Hardware Requirements
ACE-Step is designed to be accessible:
| Tier | Hardware | Experience |
|---|---|---|
| Minimum | 4GB VRAM GPU | Works, slower |
| Recommended | 8GB VRAM | Good speed |
| Optimal | Ryzen AI Max+ | Best performance |
Software Setup
ACE-Step integrates with ComfyUI via AMD's ROCm software:
Step 1: Install ROCm
# For Linux (recommended)
sudo apt update
sudo apt install rocm-dev
# Verify installation
rocminfo
Step 2: Install ComfyUI
git clone https://github.com/comfyanonymous/ComfyUI
cd ComfyUI
pip install -r requirements.txt
Step 3: Install ACE-Step Extension
cd ComfyUI/custom_nodes
git clone https://github.com/ace-step/ace-step-comfyui
pip install -r ace-step-comfyui/requirements.txt
Step 4: Download Model
# Model weights (~3GB)
huggingface-cli download ace-step/ace-step-1.5 --local-dir ./models/ace-step
Running Directly (No ComfyUI)
For optimal performance on AMD GPUs:
# Activate virtual environment
python -m venv ace-step-env
source ace-step-env/bin/activate
# Install requirements
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/rocm5.7
# Run ACE-Step
python generate.py --prompt "upbeat electronic dance track with synth lead"
Generation Modes
Text-to-Music
Simple prompt-based generation:
Create a chill lo-fi hip hop beat with soft piano and vinyl crackle
Lyrics-to-Song
Provide lyrics for vocal tracks:
[Verse 1]
Walking down the street at night
City lights are shining bright
[Chorus]
This is the moment we've been waiting for
Reference Audio
Use existing audio as a starting point:
Input: reference_song.wav
Prompt: Same style but with guitar instead of piano
Advanced Controls
ACE-Step 1.5 offers granular control:
| Control | Description |
|---|---|
| Cover generation | Create covers of existing songs |
| Repainting | Modify specific sections |
| Track separation | Split stems from generated audio |
| Multi-track | Generate multiple layers |
| Vocal-to-BGM | Convert vocals to background music |
| LoRA training | Personalize with your own style |
Comparison: Local vs Cloud
ACE-Step 1.5 vs Suno/Udio
| Aspect | ACE-Step (Local) | Suno/Udio (Cloud) |
|---|---|---|
| Cost | Hardware only | $10-20+/month |
| Generation limit | Unlimited | Quota-based |
| Internet required | ❌ No | ✅ Yes |
| Vocal quality | Good | Better |
| Speed | 30-120 sec | 10-30 sec |
| Commercial rights | ✅ Clear | ⚠️ Complex |
| Privacy | ✅ Full | ❌ Cloud processed |
When to Use Each
Choose ACE-Step if:
- Budget is a concern
- Privacy matters
- You need unlimited generations
- Working offline
- Commercial use is planned
Choose Suno/Udio if:
- Maximum vocal quality needed
- No suitable hardware
- Speed is critical
- You prefer cloud convenience
Tips for Best Results
1. Be Specific with Prompts
Weak: "A song"
Strong: "Melancholic piano ballad in the style of Ludovico Einaudi, 90 BPM, with gentle string accompaniment building to an emotional crescendo"
2. Use Style Tags
ACE-Step understands genre tags:
[Genre: Electronic]
[Mood: Euphoric]
[BPM: 128]
[Instruments: Synth, Bass, Drums]
3. Iterate in Sections
Generate verse, chorus, and bridge separately, then combine for more control.
4. Apply LoRA for Consistency
Train a LoRA on your preferred style for consistent output across multiple generations.
AMD's AI Strategy
ACE-Step support is part of AMD's broader AI push:
| Initiative | Description |
|---|---|
| ROCm | Open-source GPU compute platform |
| Ryzen AI | NPU integration in CPUs |
| MI300X | Data center AI accelerators |
| LM Studio | Local LLM framework support |
AMD is positioning itself as the "local AI" alternative to NVIDIA's cloud-focused approach.
Ryzen AI Max+ Processors
The best ACE-Step performance comes from AMD's latest:
| Processor | TOPS | Notes |
|---|---|---|
| Ryzen AI Max+ 395 | 126 | Highest consumer AI power |
| Ryzen AI 9 HX 370 | 50 | Good mobile option |
| Ryzen AI 7 340 | 39 | Entry-level |
These processors combine CPU, GPU, and NPU for AI workloads.
Key Takeaways
✅ Under 4GB VRAM - Runs on modest hardware
✅ 30 seconds for a 2-minute song on AI PRO GPU
✅ Royalty-free - Trained on non-copyrighted material
✅ MIT license - Full commercial use allowed
✅ 50+ languages - Multilingual vocal support
✅ No monthly fees - One-time hardware investment
Interested in more local AI? Check out Running Local AI on Mac and Why Your Next Laptop Might Run AI Locally.