---

name: conditioning description: "Use diffusion wrapper models conditioned on lower-rate waveforms, mel spectrograms, latents, and transform plugins." disable-model-invocation: true metadata: disco-role: operating license: MIT

Conditioning

Use this sub-skill when the task is about diffusion wrappers that condition generation on another audio representation.

Route here when the request mentions

• audio upsampling or super-resolution
• waveform to mel vocoding
• diffusion autoencoding or latent conditioning
• encoder, adapter, mel, or transform wrapper setup
• AppendChannelsPlugin or LTPlugin

Route to ../generation instead when the task is about

• base unconditional generation
• text-conditioned generation
• inpainting
• core UNet, diffusion, or sampler configuration that is not wrapper-specific

What to open first

• references/api-reference.md for constructors, prefixes, and shape contracts
• references/workflows.md for upsampler, vocoder, and autoencoder recipes
• references/troubleshooting.md for prefix, shape, and dependency fixes
• scripts/tiny_conditioning_smoke.py for a safe CPU smoke run

Quick map

• Upsampler: DiffusionUpsampler
• Vocoder: DiffusionVocoder
• Autoencoder: DiffusionAE
• Encoder contract: EncoderBase
• Adapter hooks: AdapterBase
• Mel helper: MelSpectrogram
• Transform helpers: AppendChannelsPlugin, LTPlugin

The bundled smoke uses a local dummy encoder and stays away from optional external encoder packages.