---
title: Fragmenta
emoji: 🎵
colorFrom: purple
colorTo: blue
sdk: docker
app_port: 7860
pinned: false
license: agpl-3.0
---

# Fragmenta — AI Audio Generation & Fine-Tuning

Generate and fine-tune audio from text prompts using Stable Audio 3.

Fragmenta is licensed under AGPL-3.0. **Powered by Stability AI** — the Stable
Audio 3 model weights are used under the Stability AI Community License.

**Hardware:** cpu-basic (CPU)

## Getting Started

1. Download an SA3 checkpoint via the in-app Checkpoint Manager, or place one
   under `models/pretrained/sa3/hub/` in the Space Files tab.
   - `sa3-small-music` (recommended for CPU Spaces)
   - `sa3-medium` (recommended for GPU Spaces with Flash Attention 2)
2. The Space will auto-rebuild after the upload.
3. Use the **Data Processing** tab to upload audio + prompts.
4. Use the **Training** tab to fine-tune.
5. Use the **Generation** tab to generate audio.