Text-to-music
describe a song, get audio
// official site: github.com ↗
AUDIO & MUSIC · PRO TIER
MusicGenpro
MusicGen is Meta AI's text-to-music model (part of AudioCraft) — generates 30+ seconds of stereo music from text prompts or melody conditioning. Three model sizes (300M / 1.5B / 3.3B params) for VRAM/quality tradeoff. Genre, instrument, mood, BPM all controllable via prompts.
// What it is
A closer look.
MusicGen is Meta AI's text-to-music model (part of AudioCraft) — generates 30+ seconds of stereo music from text prompts or melody conditioning. Three model sizes (300M / 1.5B / 3.3B params) for VRAM/quality tradeoff. Genre, instrument, mood, BPM all controllable via prompts.
The open answer to Suno / Udio for music generation, with full control over the entire pipeline.
// Use cases
What it's for.
Concrete scenarios where teams pick MusicGen over the SaaS alternative.
Melody-conditioned generation
hum a tune, get a full instrumental
Custom soundtracks
for video, podcast, game, app
Multiple genres
orchestral, electronic, jazz, folk, classical, hip-hop
BPM + instrument control
via prompts
Royalty-free music generation
(with appropriate license use)
// Who it's for
Built for these teams.
If your team profile matches one of these, MusicGen is a strong fit out of the box.
Video creators
generating royalty-free soundtracks
Podcasters
producing intro/outro themes
Game developers
prototyping music for levels
Marketing teams
producing campaign-specific audio
App developers
generating ambient/UX audio
Hosting providers
selling music gen tier
// Differentiators
Why teams pick MusicGen.
When evaluating self-hosted options for this category, here are the dimensions on which MusicGen consistently lands above the alternatives.
- ✓MIT license — (code); model weights have research/commercial separation
- ✓Strong quality — for open-source music gen
- ✓Multiple sizes — choose VRAM vs quality tradeoff
- ✓Melody conditioning — unique feature, not in Suno open versions
- ✓Meta backing — AudioCraft = part of well-funded Meta AI research
- ✓Long-form support — chain 30-sec segments for longer tracks
// Integrations
Connects to.
The stack you'll plug MusicGen into — services, protocols, and adjacent apps in the BluixApps catalog.
◇
Gradio web UI
out of box
◈
HuggingFace Transformers
pipeline
◆
AudioCraft library
Python API for batch
▣
Pair with Demucs
stem-separate generated tracks for remix flexibility
▦
Pair with video AI
generate B-roll → match with MusicGen score
// Adoption & deployment
Notable users & community
- 21k+ GitHub stars (parent AudioCraft repo)
- Meta AI corporate backing
- Used in commercial video production pipelines
- Active fine-tuning community for specific genres
- Featured in major AI tracker newsletters
What we ship
- Cloned
facebookresearch/audiocraftrepo - pytorch/pytorch CUDA 12.4 base + ffmpeg
- musicgen_app Gradio launcher
- Persistent volumes: repo, models (~5 GB), output
- Port 7868 mapped
- Install report at
/root/bluixapps/musicgen.txt - License caveat clearly documented (CC-BY-NC for outputs)
- Pairing suggestions (Demucs for remix, video AI for B-roll match)
- GPU pre-flight check via
bluixapps_ensure_nvidia_runtime - Backup hook covers models + outputs
// Tips & operations
Run it properly.
Operational guidance from running this in production — what to lock down, what surprises people.
// PERFORMANCE
Model size
// SECURITY
Prompts
mention genre, instruments, BPM, mood explicitly
// OPERATIONS
Length
30 seconds per pass; chain for longer
// RELIABILITY
Sample
"Upbeat acoustic guitar folk with cheerful violin solo, 90 BPM"
// DEPLOYMENT
VRAM
scale model to available; medium is sweet spot
// SCALING
License check
original Meta weights have research-only clauses for full models;
12288
// min ram (MB)
20
// min disk (GB)
7868
// access port
http
// protocol
pro
// bluixapps tier
// Alternatives in Audio & music
Compare with
Project resources
Official sitegithub.com ↗