ryanbw/musicgen-medium-jazz-soul-lora

MusicGen Medium Jazz-Soul LoRA (V17)

This model is a fine-tuned LoRA adapter for facebook/musicgen-medium, specialized in generating high-quality Jazz, Soul, and R&B music with strong structural consistency and long-form stability.

Model Details

Model Description

• Developed by: ryanbw
• Model type: Audio Generation (LoRA Adapter)
• Language(s): English (Prompts)
• License: Apache 2.0
• Finetuned from model: facebook/musicgen-medium

Model Sources

• Repository: https://github.com/bowenwanryan-alt/musicgen-medium-jazz-soul-lora

Uses

Direct Use

This model is designed to generate Jazz and Soul music clips. It excels at creating structured loops (20-25 seconds ) with clear instrumental separation, specifically piano-driven jazz and rhythmic R&B.

How to Get Started with the Model

Use the code below to load the model and generate music:

import torch
from transformers import MusicgenForConditionalGeneration, AutoProcessor
from peft import PeftModel

model_id = "facebook/musicgen-medium"
lora_id = "ryanbw/musicgen-medium-jazz-soul-lora" 

model = MusicgenForConditionalGeneration.from_pretrained(model_id, torch_dtype=torch.float16).to("cuda")
model = PeftModel.from_pretrained(model, lora_id)
processor = AutoProcessor.from_pretrained(model_id)

# Example Prompt
prompt = "A professional recording of Bossa Nova rhythm, Latin Jazz influence, 129 BPM, Key of D, smooth groove, studio quality, structured musical loop."
inputs = processor(text=[prompt], return_tensors="pt").to("cuda")

# Generate ~20 seconds of audio
audio = model.generate(**inputs, max_new_tokens=1000, guidance_scale=3.5, temperature=0.85)