MeanAudio: Fast and Faithful Text-to-Audio Generation with Mean Flows

MeanAudio is a novel text-to-audio generator that uses MeanFlow to synthesize realistic and faithful audio in few sampling steps. It achieves state-of-the-art performance in single-step audio generation and delivers strong performance in multi-step audio generation.

📖 Arxiv | 💻 GitHub | 🤗 Model | 🚀 Space | 🌐 Project Page

Prompt

Duration

1 30

Guidance Scale

1 10

Sampling Steps

1 25

Model Variant

Seed

1 100

🎵 Audio Sample

Prompt Used

Examples

Prompt	Duration	Guidance Scale	Sampling Steps	Model Variant	Seed