Stable-Audio-Demo (stability-ai.github.io)

Article summary

The Stable Audio demo showcases a model that can generate variable-length and long-form stereo music at 44.1kHz, as well as sound effects. The model can produce high-quality audio, but some listeners note that the music sounds unnatural or 'off' in certain ways. The demo includes comparisons with state-of-the-art models and examples of autoencoder reconstructions. The model's capabilities are demonstrated through various prompts and audio examples.

Main themes

• AI music generation
• Sound effects generation
• Text prompting limitations
• Human creativity and AI
• Intellectual property concerns
• Music production constraints

What commenters say

• The generated music sounds unnatural or 'off' due to the model's lack of understanding of human music production constraints.
• The model's ability to generate sound effects is more impressive and has potential applications in areas like indie game development.
• Text prompting is a coarse tool for generating music and may not be the most effective way to produce high-quality output.
• The use of AI in music generation can be seen as a tool to support human creativity, rather than replace it.
• The lack of public weights and datasets for the model may be due to intellectual property concerns.
• Some commenters believe that the model's potential is limited by its reliance on text prompting and that other input methods, such as MIDI or control nets, may be more effective.