arioso.platforms.stable_audio_25.adapter#

Stable Audio 2.5 adapter (fal.ai): text-to-audio and audio-to-audio.

Unlike the local stable_audio platform (Stable Audio Open via diffusers, which has no strength knob), the hosted 2.5 audio-to-audio endpoint takes a strength in [0, 1]: low values keep the input nearly intact, high values let the prompt take over. That makes it the arioso backend for “the same piece, subtly changed”.

Classes

Adapter(config)

fal.ai Stable Audio 2.5 adapter.
class arioso.platforms.stable_audio_25.adapter.Adapter(config)[source]#

Bases: object

fal.ai Stable Audio 2.5 adapter.

generate(prompt, *, duration=None, num_steps=8, guidance=1.0, seed=None, audio_input=None, audio_input_strength=0.8, fetch=True, **kwargs)[source]#

Generate audio from a prompt, or transform audio_input toward it.

Parameters:

  • prompt (str) – Text description of the desired audio (required).

  • duration (float) – Output length in seconds (1-190). Text-to-audio defaults to 30 s (fal’s own default is its 190 s maximum, the most expensive clip); audio-to-audio defaults to the input’s length.

  • num_steps (int) – Denoising steps (num_inference_steps, default 8).

  • guidance (float) – Prompt adherence (guidance_scale, default 1).

  • seed (int) – Random seed for reproducibility.

  • audio_input – Audio to transform (path, bytes, Song, URL, …). When given, the audio-to-audio endpoint is used.

  • audio_input_strength (float) – Denoising strength for audio-to-audio (0-1, default 0.8); lower keeps more of the input.

  • fetch (bool) – Download the result so audio_bytes is populated.

Return type:

Song

Returns:

A completed Song (audio_bytes and audio_url).