gary4beatbox 2.0

Gary processes your input audio with remarkable fidelity, providing a versatile tool for audio transformation and continuation.

The initial iteration was primarily designed to extend beatbox sequences. However, the latest version expands its capabilities significantly through integration with Terry, allowing for broader use cases.

You may record audio directly via your microphone, with options for noise cancellation and a count-in feature if needed. Once recorded, simply send your audio to Gary for processing.

Gary can either continue from the point where your input ends or crop the tail of its previous output to commence anew. If the generated output does not meet your expectations, you can retry with different settings to achieve a different result from the same starting point.

Notable features include:

• Ability to select any Audiocraft or MusicGen model available on the Hugging Face hub by entering the model name in the settings.
• The 'Tame the Gary' mode aims to focus output on percussion and drum elements, potentially enhancing beatbox continuity. Its effectiveness is experimental and appears most compatible with the 'thepatch/vanya_ai_dnb_0.1' model, although results may vary.
• The 'prompt duration' parameter specifies the number of seconds of audio to be utilized. When activating 'send to Gary,' this duration is counted from the beginning of your input; when using 'continue' or 'retry,' it is measured from the end of the current audio clip.
• You can share your creations as videos for platforms like Twitter or toggle this option to share only the audio content.

The app now includes 'Terry,' a new model developed by Meta called MelodyFlow. Terry can transform both your recorded audio and outputs generated by Gary. You may undo recent modifications and reapply transformations, adjusting the 'flowstep' parameter—akin to the transformation's intensity—for optimal shaping. Generally, higher values tend to better preserve the original form of your input.

To personalize Terry’s transformations, you can choose preset prompts or input custom prompts—an experimental feature not available in other applications. More information about Terry is available at https://huggingface.co/spaces/facebook/Melodyflow.

As this project is open source and free, there is uncertainty regarding the longevity of our backend infrastructure. If you have technical expertise, you may replace our backend URL by building from source, available at https://github.com/betweentwomidnights/gary-backend-combined. This backend shares its core with Gary4Live, a free Ableton plugin capable of versatile processing.