README.md

metadata

title: Sonisphere
emoji: 🐢
colorFrom: green
colorTo: gray
sdk: gradio
sdk_version: 5.20.0
app_file: app.py
pinned: false

Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference

Sonisphere Demo

This is a Hugging Face Spaces demo for MMAudio, a powerful model for generating realistic audio for videos.

🎥 Features

• Upload any video and generate matching audio
• Control the generation with text prompts
• Adjust generation parameters like steps and guidance strength
• Process videos up to 30 seconds in length

🚀 Usage

1. Upload a video or use one of the example videos
2. Enter a text prompt describing the desired audio
3. (Optional) Add a negative prompt to specify what you don't want
4. Adjust the generation parameters if needed
5. Click "Submit" and wait for the generation to complete

⚙️ Parameters

• Prompt: Describe the audio you want to generate
• Negative prompt: Specify what you don't want in the audio (default: "music")
• Seed: Control randomness (-1 for random seed)
• Number of steps: More steps = better quality but slower (default: 25)
• Guidance Strength: Controls how closely the generation follows the prompt (default: 4.5)
• Duration: Length of the generated audio in seconds (default: 8)

📝 Notes

• Processing high-resolution videos (>384px on shorter side) takes longer and doesn't improve results
• The model works best with videos between 5-30 seconds
• Generation time depends on video length and number of steps

🔗 Links

• Project Page
• GitHub Repository
• Paper

📜 License

This demo uses the MMAudio model which is released under the MIT license.