These models generate and modify music from text prompts and raw audio. They combine large language models and diffusion models trained on text-music pairs to understand musical concepts.
Featured models

minimax/music-1.5
Music-1.5: Full-length songs (up to 4 mins) with natural vocals & rich instrumentation
Updated 1 month, 2 weeks ago
12K runs


stability-ai/stable-audio-2.5
Generate high-quality music and sound from text prompts
Updated 1 month, 2 weeks ago
4K runs


meta/musicgen
Generate music from a prompt or melody
Updated 1 year, 7 months ago
3.1M runs
Recommended Models
If speed is your main concern, models like lucataco/ace-step are designed for fast generation of longer tracks. Larger or more complex models such as meta/musicgen (3.5 B parameters) tend to take longer and cost more per minute of audio.
For a good balance, the medium- or small-sized variants of meta/musicgen (for example, the “melody” version) are strong choices: they provide solid audio quality without the heavy compute of the largest models. Models like minimax/music-1.5 and google/lyria-2 offer higher fidelity and structure but at a higher cost.
If you want a structured song with vocals and instrumentation, minimax/music-1.5 is designed for full-length tracks with vocals, verse-chorus structure, and rich arrangements. google/lyria-2 delivers professional-grade stereo audio but may have shorter duration limits. Smaller meta/musicgen models can work well for instrumental tracks.
For background or looped music, you don’t need the complexity of a full-song model. A loop-optimized variant, such as andreasjansson/musicgen-looper, can generate fixed-BPM loops more quickly and at lower cost. This is ideal for short, repeatable segments used in videos or games.
If you need vocals and a polished result, start with minimax/music-1.5 or google/lyria-2. For instrumental sketches or quick iterations, use a mid-tier meta/musicgen variant. If you already have a chord progression, pick a chord-conditioned model like TODO: add link for "MusicGen-Chord". Experiment with prompts and durations to get the exact style you want.
There are a few core approaches to how these models create music:
Most models output stereo audio at 32 kHz or 48 kHz. Duration limits vary—some meta/musicgen models focus on short clips, while models like minimax/music-1.5 support up to around four minutes. Higher-fidelity models produce more polished instrumentation and vocals.
You can fine-tune models like meta/musicgen with your own audio dataset and deploy them to Replicate. Fine-tuning requires some setup and compute, especially if you want to maintain or add vocal output.
Yes, many users use these models in commercial projects. However, you should check the model’s license and any underlying data usage restrictions. Pay special attention to vocal output, which may involve more complex rights considerations.
Running a model typically involves providing a text prompt and optional inputs like melody or chords. The model generates an audio clip you can download. Each model has its own set of input fields, so check the model’s documentation before running.
Start with short clips to test styles and costs before committing to longer tracks. Use chord-conditioned models if you already have a harmonic structure. Loop models are best for game audio or UX sound design. Even with AI, you may want to do some final mixing or mastering for professional use.
Recommended Models

minimax/music-01
Quickly generate up to 1 minute of music with lyrics and vocals in the style of a reference track
Updated 1 month, 2 weeks ago
405.7K runs

google/lyria-2
Lyria 2 is a music generation model that produces 48kHz stereo audio through text-based prompts
Updated 1 month, 2 weeks ago
36.1K runs


lucataco/ace-step
A Step Towards Music Generation Foundation Model text2music
Updated 5 months, 2 weeks ago
60.9K runs


zsxkib/flux-music
🎼FluxMusic Text-to-Music Generation with Rectified Flow Transformer🎶
Updated 1 year, 1 month ago
8.7K runs


sakemin/musicgen-remixer
Remix the music into another styles with MusicGen Chord
Updated 1 year, 9 months ago
17.7K runs


lucataco/magnet
MAGNeT: Masked Audio Generation using a Single Non-Autoregressive Transformer
Updated 1 year, 9 months ago
2.8K runs


sakemin/musicgen-stereo-chord
Generate music in stereo, restricted to chord sequences and tempo
Updated 1 year, 11 months ago
3.3K runs


sakemin/musicgen-chord
Generate music restricted to chord sequences and tempo
Updated 1 year, 11 months ago
2.9K runs


fofr/musicgen-choral
MusicGen fine-tuned on chamber choir music
Updated 2 years ago
4.7K runs


andreasjansson/musicgen-looper
Generate fixed-bpm loops from text prompts
Updated 2 years, 4 months ago
59.5K runs


riffusion/riffusion
Stable diffusion for real-time music generation
Updated 2 years, 10 months ago
1.1M runs