Woosh-DFlow

Sound prompt

Examples

Duration

Clip length in seconds.

3 10

Takes

Generate one or two variations per request.

1 2

Prompt strength

Higher values follow the prompt more tightly.

0 9

Seed

Use -1 for a random seed.

Take 1

The first request may wait while the official DFlow, autoencoder, and text-conditioning checkpoints are downloaded and loaded.

Model weights are downloaded from the official SonyResearch/Woosh v1.0.0 release: Woosh-DFlow, Woosh-AE, and TextConditionerA. The released weights are licensed CC-BY-NC; the upstream inference code is MIT/Apache-2.0.