conda install bangers

Local-first AI music generation studio

Local AI Music Studio to Generate and remix music entirely on your own machine. Built on ACE-Step 1.5.

Use it to generate, remix, save, and play AI music on your own machine. The app includes text-to-music, custom generation, remix mode, AI DJ chat, radio stations, a library, and a full audio player.

Quick Start

Prerequisites

Git
mise installed and activated in your shell

The repo pins Python 3.11, Node.js 20, pnpm 9.15.9, and conda in .mise.toml.

Run Locally

git clone https://github.com/DEKHTIARJonathan/conda-install-bangers.git
cd conda-install-bangers
mise install
mise run setup
mise run dev

The launcher starts:

backend: https://localhost:8000
frontend: https://localhost:3000
runtime data: backend/data/
model cache: .cache/models/

The dev server uses a self-signed HTTPS certificate. Your browser will warn the first time; proceed to the local site. Press Ctrl+C to stop both servers.

On first launch, no model is loaded. Open the Models page, download/select a DiT model, and optionally select an ACE language model and a chat LLM. Selections are stored in backend/data/conda-install-bangers.db and restored on restart.

Daily Commands

mise run dev        # Start backend and frontend
mise run test       # Run backend and frontend tests once
mise run clean      # Reset local DB/audio/uploads, keep downloaded models

Launcher flags:

python start.py --install   # Force dependency reinstall
python start.py --no-open   # Do not auto-open the browser

After pulling updates:

git pull
mise run setup
mise run dev

Models

All downloads and active selections happen in the Models page. Model selection is intentionally not controlled by environment variables.

Current ACE model registry:

Type	Models
DiT	`acestep-v15-turbo`, `acestep-v15-base`, `acestep-v15-sft`, `acestep-v15-turbo-continuous`, `acestep-v15-xl-turbo`
ACE language model	`acestep-5Hz-lm-1.7B`, `acestep-5Hz-lm-0.6B`, `acestep-5Hz-lm-4B`, or no LM
Chat LLM, MLX	`Qwen3-0.6B-4bit`, `Qwen3-1.7B-4bit`, `Qwen3-4B-4bit`, `Qwen3-8B-4bit`
Chat LLM, Transformers	`Qwen3-1.7B`, `Qwen3-4B-Instruct-2507`, `Qwen3-8B-FP8`, `Qwen3-14B-FP8`, `Qwen3-30B-A3B-Instruct-2507-FP8`

Disk usage depends on what you download. The default ACE bundle is about 10 GB, the XL DiT is about 20 GB, and larger chat LLMs add more.

Rough ACE LM guidance:

VRAM	Suggested LM
<=6 GB	none
6-8 GB	`acestep-5Hz-lm-0.6B`
8-16 GB	`acestep-5Hz-lm-1.7B`
16-24 GB	`acestep-5Hz-lm-1.7B` or `acestep-5Hz-lm-4B`
>=24 GB	`acestep-5Hz-lm-4B`

Configuration

Common environment variables:

Variable	Default	Purpose
`BANGERS_HOST`	`0.0.0.0`	Backend bind address
`BANGERS_PORT`	`8000`	Backend port
`BANGERS_DEVICE`	`auto`	`auto`, `cuda`, `mps`, or `cpu`
`BANGERS_LM_BACKEND`	`mlx` on macOS, `nano-vllm` elsewhere	ACE LM backend
`BANGERS_AUDIO_FORMAT`	`flac`	Default output format
`BANGERS_BATCH_SIZE`	`2`	Default samples per generation
`BANGERS_INFERENCE_STEPS`	`8`	Default DiT steps
`BANGERS_GUIDANCE_SCALE`	`7.0`	Default guidance scale
`BANGERS_THINKING`	`true`	Default 5 Hz LM thinking mode
`BANGERS_DATA_DIR`	`backend/data`	SQLite DB, audio, uploads
`BANGERS_MODEL_CACHE_DIR`	`.cache/models`	Model/cache root
`ACESTEP_PROJECT_ROOT`	`.cache/models`	ACE checkpoints and chat LLM root

Most generation defaults can also be changed in the app under Settings.

Production

The current Docker Compose stack targets a single NVIDIA Linux host and serves HTTP:

cp .env.example .env
docker compose build
docker compose up -d

Open http://localhost:3000.

Compose uses two named volumes:

bangers-data: SQLite DB, generated audio, uploads
bangers-models: model weights and Hugging Face cache

See DEPLOY.md for GPU prerequisites, upgrades, backups, and failure checks.

Development

See DEVELOPMENT.md for local setup, TLS behavior, cache paths, and test commands.

Run tests:

mise run test

Or run each side directly:

(cd backend && conda run --prefix .conda pytest -v)
pnpm --dir frontend exec vitest --run

Credits and License

Built on ACE-Step 1.5. See ATTRIBUTION.md for community inspirations.

Licensed under the MIT License.

isVoid/conda-install-bangers

README