108 lines
3.4 KiB
Plaintext
108 lines
3.4 KiB
Plaintext
==============================================================
|
|
HOT-Step 9000 — High-Performance AI Music Generation
|
|
==============================================================
|
|
|
|
QUICK START
|
|
-----------
|
|
1. Extract this folder anywhere you like.
|
|
2. Double-click "HOT-Step.bat" to start.
|
|
3. Your browser will open to http://localhost:3001 automatically.
|
|
4. On first launch, go to Settings > Model Manager to download
|
|
the AI models (~7 GB). You'll need an internet connection.
|
|
|
|
That's it! No installation required.
|
|
|
|
|
|
REQUIREMENTS
|
|
------------
|
|
- Windows 10/11 (64-bit)
|
|
- NVIDIA GPU with recent drivers (RTX 2060 or newer recommended)
|
|
* CPU-only mode works but is significantly slower
|
|
- ~10 GB free disk space (for models + generated audio)
|
|
- Internet connection for first-run model downloads
|
|
|
|
|
|
GPU SUPPORT
|
|
-----------
|
|
CUDA Build (NVIDIA):
|
|
Supports all NVIDIA GPUs with recent drivers:
|
|
- RTX 2000 series (Turing)
|
|
- RTX 3000 series (Ampere)
|
|
- RTX 4000 series (Ada Lovelace)
|
|
- RTX 5000 series (Blackwell)
|
|
|
|
Vulkan Build (AMD / NVIDIA / Intel):
|
|
Uses Vulkan for GPU acceleration. Works with most modern GPUs
|
|
that have Vulkan 1.1+ support. Requires GPU drivers with the
|
|
Vulkan runtime installed (included with most driver packages).
|
|
|
|
IMPORTANT: Use the 4B LM model with the Vulkan build.
|
|
The 1.7B LM model produces corrupted output due to precision
|
|
issues in the Vulkan compute shaders. The 4B model works
|
|
correctly. This does not affect the DiT or VAE pipelines.
|
|
|
|
CPU Build:
|
|
Runs entirely on CPU. No GPU required, but generation will be
|
|
significantly slower.
|
|
|
|
Make sure your GPU drivers are up to date:
|
|
https://www.nvidia.com/en-us/drivers/
|
|
|
|
|
|
STEM SEPARATION (Optional)
|
|
---------------------------
|
|
Stem separation requires additional runtime files (~1.3 GB).
|
|
When you first use Stem Studio, you'll be prompted to download
|
|
them from the Model Manager.
|
|
|
|
These files include ONNX Runtime and cuDNN libraries required
|
|
for the neural network stem separator.
|
|
|
|
|
|
CONFIGURATION
|
|
-------------
|
|
To customize settings, copy ".env.example" to ".env" and edit it.
|
|
Most settings can also be changed from the Settings page in the app.
|
|
|
|
|
|
FOLDER STRUCTURE
|
|
----------------
|
|
HOT-Step.bat — Launch the application
|
|
runtime/ — Node.js runtime (do not modify)
|
|
engine/ — C++ inference engine
|
|
server/ — Application server
|
|
ui/ — Web interface
|
|
models/ — AI model files (downloaded on first run)
|
|
adapters/ — LoRA adapters (optional)
|
|
Essentia/ — Audio analysis tool
|
|
noise_samples/ — Noise profiles for denoising (optional)
|
|
|
|
|
|
TROUBLESHOOTING
|
|
---------------
|
|
Q: The app won't start.
|
|
A: Make sure you extracted the FULL zip. Don't move individual
|
|
files. Try running HOT-Step.bat from an elevated command prompt.
|
|
|
|
Q: Generation is very slow.
|
|
A: Check that your NVIDIA drivers are installed. The app will
|
|
fall back to CPU if no GPU is detected. See Settings > Health
|
|
for GPU status.
|
|
|
|
Q: No sound / playback issues.
|
|
A: Make sure your browser allows audio playback. Try a different
|
|
browser (Chrome or Edge recommended).
|
|
|
|
Q: Model download fails.
|
|
A: Check your internet connection. Downloads can be resumed —
|
|
just click the download button again.
|
|
|
|
|
|
LICENSE & CREDITS
|
|
-----------------
|
|
Built on ACE-Step (MIT License) — https://github.com/ace-step
|
|
GGML inference framework — https://github.com/ggml-org/ggml
|
|
|
|
For support and updates:
|
|
https://github.com/scragnog/HOT-Step-CPP
|