this is the official vst/au for gary.

version: 2.0 (11.12.2025 update)

gary used to be a max4live device with 3 open source music models inside of it. it was kinda a PITA to use...

now, it's a vst with FOUR models inside it so that fl studio, reaper, bitwig, and even garageband users can play.

what else is new in v2.0?

- linux support! (tested with bitwig studio so far)

- magenta-realtime model added. i named it darius.

- use custom stable-audio-open-small finetunes.

how it works

1. put it on your master track.

2. press play in your daw (with layers solo'd if you want. i do this alot with percussion)

3. save your recorded buffer.

gary tab: use the first x seconds of the buffer you recorded and 'send to gary' to have it iterate on the input audio. once it's in the output waveform, you can crop it/continue it using the last x seconds of audio as input.

gary uses musicgen finetunes: audiocraft repo

jerry tab: generate bpm aware loops of either drums/instruments or 'auto' that you can drag into your daw, or have the other two models iterate on it. (this model generates in under a second!)

- now supporting any SAOS finetunes hosted on huggingface

jerry is stable-audio-open-small: SAOS on huggingface , stable-audio-tools repo

terry tab: transform either your recorded audio or the output waveform according to presets or a custom prompt you type (up to around 40 seconds)

terry is meta's melodyflow model: melodyflow on huggingface

darius tab: iterate on your input audio or other model outputs with 48khz stereo using either the base model or your custom finetunes. requires separate backend (using our huggingface space or local docker containers if you've got the gpu).

darius is google's magenta-realtime model: magenta-realtime repo

learn more about it on youtube

i'll add more music models as i see fit. this plugin is a playground for them with a focus on input audio/collaboration.

if you have a suggestion for a model you'd like to see inside the vst, give me a holler at kev@thecollabagepatch.com or hop in discord

i run a backend free of charge so that everyone can use it, but inside you'll also find an installer so that you can run the backend yourself locally if you have ~10+ gb of gpu vram.

there's also some stuff about finetuning models inside here. if you throw a finetune up on huggingface (musicgen, stable-audio-open-small or magenta-realtime), it will run inside here.