README.md

---
title: Stem-0 DEV (r2c checkpoints)
emoji: 🎚️
colorFrom: blue
colorTo: indigo
sdk: gradio
python_version: "3.12"
sdk_version: 6.2.0
app_file: app.py
startup_duration_timeout: 1h
pinned: false
short_description: Generate the missing stem that fits your audio
---

# Stem-0 · missing-stem generation

Give it an audio **context** (e.g. drums+bass, or a full instrumental) and it generates the
**missing stem** so that it fits — tempo, key, arrangement. For vocals you supply the lyrics.

Base model: `ACE-Step/acestep-v15-xl-base` · LoRA: `fcolooo/stem-0-r128` (r128, context-contrastive recipe).

**ZeroGPU note:** the first generation loads ~20 GB of weights onto the GPU and is slow; later ones are fast.
Keep `steps` low while testing — GPU quota is per-day.