Enum StableAudioModelSize

Namespace

AiDotNet.Audio.StableAudio

Assembly

AiDotNet.dll

Specifies the size variant of the Stable Audio model.

public enum StableAudioModelSize

Fields

Base = 1

Base model variant (800M parameters). Default choice.

• T5 encoder: 768 hidden dim
• DiT: 1024 hidden dim, 24 blocks
• Good balance of quality and speed

Large = 2

Large model variant (1.5B parameters).

• T5 encoder: 1024 hidden dim
• DiT: 1536 hidden dim, 32 blocks
• Highest quality, requires significant GPU memory

Open = 3

Stable Audio Open variant.

• Open-source model with permissive license
• Optimized for music generation
• Based on Base architecture

Small = 0

Small model variant (300M parameters).

• T5 encoder: 256 hidden dim
• DiT: 512 hidden dim, 12 blocks
• Fast inference, suitable for experimentation

V2 = 4

Stable Audio 2.0 variant.

• Improved architecture with better coherence
• Extended duration support (up to 3 minutes)
• Enhanced stereo output

Remarks

Stable Audio is a latent diffusion model by Stability AI for high-quality audio generation. It uses a Diffusion Transformer (DiT) architecture instead of U-Net for improved quality and supports variable-length audio generation.

For Beginners: Think of model sizes like different quality levels:

• Small: Fast generation, good for experimentation (300M parameters)
• Base: Balanced quality and speed (800M parameters)
• Large: Best quality, requires more resources (1.5B parameters)
• Open: Open-source variant with permissive license

Start with Small for testing, use Large for production.