README.md

metadata

license: mit
language: en
pipeline_tag: text-to-audio
base_model: hkchengrex/MMAudio

Fine-tuned MMAudio Model

This repository contains a fine-tuned version of the MMAudio model, specializing in generating audio from video, image, and text prompts. The model was fine-tuned on a private dataset, resulting in the cloud19/NSFW_MMaudio checkpoint. Specializing in NSFW videos of various kinds.

Original Project: hkchengrex/MMAudio

Model Details

• Base Architecture: large_44k (from the original MMAudio)
• Fine-tuning: The model was fine-tuned on a private dataset of NSFW content to explore the model's adaptability to specific audio domains. Due to the nature of the training data, the generated audio may reflect mature themes.
• Capabilities: Video-to-Audio, Image-to-Audio, Text-to-Audio.

How to Use

This repository includes a fully functional Gradio (app.py). To run it locally, follow these steps. GPU 12-16 GB preferred.

Prerequisites:

• Python 3.10+
• Git and Git LFS (git-lfs must be installed)

Step-by-Step Instructions

1. Clone the repository: Open your terminal and clone the repository. git-lfs will automatically handle downloading the large model file.

   git clone https://huggingface.co/cloud19/NSFW_MMaudio
   cd NSFW_MMaudio
   

2. Set up a virtual environment (recommended):

   python -m venv venv
   source venv/bin/activate
   # On Windows, use: venv\Scripts\activate
   

3. Install the required dependencies: The requirements.txt file contains all necessary packages.

   pip install -r requirements.txt
   

4. Run the Gradio application: This will launch a local web server.

   python app.py
   

Now you can use the web interface to generate audio with your fine-tuned model.

Limitations:

• The model's output is heavily influenced by its training data.
• Content Warning: Due to the NSFW nature of the fine-tuning dataset, the model may generate explicit or mature audio content. User discretion is advised.
• Like all generative models, it may produce artifacts or unexpected results.

Original Model Citation

All credit for the original architecture, code, and pre-trained models goes to the authors of MMAudio. Please refer to their repository for more details.

• Original GitHub: https://github.com/hkchengrex/MMAudio