OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
Audio & Voice AI

Tangoflux

TangoFlux is a text-to-audio generation model that uses flow matching to produce audio conditioned on a text prompt and duration. The project provides command-line and Python inference examples, a local web interface, a Hugging Face demo and a ComfyUI custom-node option.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Tangoflux — OpenViST project profile in Audio & Voice AI.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use TangoFlux when you need prompt-conditioned audio clips of 1–30 seconds and want an inference route ranging from a demo to Python or command-line integration.

Who it is for

  • Developers generating audio from text prompts in Python or from a command line.
  • People who want to try the model through its web demo or local web interface.
  • Researchers and model developers working with audio generation or the documented training pipeline.

Documented uses

  • Generate audio from a written description and save it as a WAV file.
  • Try text-to-audio generation through a demo or local interface.
  • Train or adapt the model using prepared audio data and the documented Accelerate workflow.
Before you choose

Requirements & limits

The repository provides a Hugging Face Space demo and a Colab notebook. Its instructions also describe installing the package from GitHub to run the local interface, CLI or Python examples. The repository states that commercial model use requires registration with Stability AI or a separate commercial license.

  • Generated clips are limited to 1–30 seconds in the documented inference instructions.
  • The Python API requires a duration argument; the repository’s example saves audio at 44.1 kHz.
  • The project reports different trade-offs between 25 and 50 inference steps: 25 steps are described as faster with similar quality, while 50 steps yielded the best results in its evaluation.
  • Training requires prepared audio and configuration files; multi-GPU training uses Hugging Face Accelerate.
  • The repository specifies Stability AI attribution and license terms, and separate restrictions for WavCaps-sourced data.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Generate a WAV file from a text prompt with Python

Use the documented Python inference example to generate and save a short audio clip.

Before you start

  • Install TangoFlux from the GitHub repository with pip.
  • Python packages torchaudio and tangoflux.
  • A text prompt and a duration between 1 and 30 seconds.

Steps

  1. Install TangoFlux using the repository’s pip command.

  2. Create a TangoFluxInference instance with the model name declare-lab/TangoFlux.

  3. Call model.generate with a text prompt, a duration and a step count.

  4. Save the returned audio with torchaudio.save at 44,100 Hz.

Expected outcome

A generated audio waveform saved as a WAV file.

Guide limitations

  • The documented Python example requires a duration between 1 and 30 seconds.
  • The repository gives an example using 50 steps but does not establish that this setting suits every use case.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
18 October 2026
Original website title
declare-lab/TangoFlux · Official homepage title · retained in its source language
How we review sourcesSuggest a correction