​ABOUT THE WORKSHOP

​🎚️ Audio Flows: Train Your Own Real-Time Generative Audio Model w/ CoDiCodec-Flow

​🗓️ 10 June 2026 · 16:00–18:00 CEST · Online · Free

​A hands-on introduction to training and performing with your own generative audio model, using CoDiCodec-Flow — an open-source pipeline that learns to improvise continuations of any sound you feed it. Think of it as teaching a model your sonic vocabulary, then jamming with it live.

​Who it's for

​Musicians, sound artists, researchers, and curious technologists interested in machine learning for audio.

​Comfort with the command line and a basic familiarity with Python helps, but no prior ML experience is required - we'll explain the concepts as we go.

​What we'll do

• ​Get a feel for the architecture: how the model compresses sound into a compact latent space, and how it learns to generate inside that space.

• ​Prepare a small personal dataset of audio — your own recordings, field recordings, samples, or a favourite album.

• ​Pre-encode it and launch a training run on your laptop.

• ​Sample from intermediate checkpoints and listen to how the model evolves as it trains.

• ​Drive the model in real time: stream chunk-by-chunk generation straight to the speakers, reseed and re-prompt live, and explore the latent-space controls (per-channel bias, summary-token scaling, LFO-driven parameter motion).

• ​Discuss what's interesting and unsettling about training small generative models on personal corpora — authorship, dataset intimacy, the aesthetics of "underfitted" models.

​What you'll leave with

• ​Knowledge on how to train and generate with your own CoDiCodec-Flow model.

• ​A small custom-trained model checkpoint — won't be production-quality after one workshop, but enough to hear your dataset starting to emerge from noise.

• ​Audio examples from offline sampling and live improvisation.

• ​A real-time setup you can keep developing afterward as an instrument.

​What we'll ponder

​What does it mean to train a model on a body of work you love? How does chunked, causal generation feel different from prompt-and-render workflows? Where is the instrument — in the weights, the prompt, the sampler, or the human hands on the keyboard?

​What to come with

• ​We’ll be using Google Colab for the training process explanation. If you have an M-series Mac or GPU enabled PC, you can follow along locally

• ​Comfort opening a terminal and running basic commands (or dare trying)

• ​~20–60 mins of audio you'd like to train on (your own recordings, field recordings, samples, or a favourite album) in .Wav format

• ​Headphones

​ABOUT MOISÉS HORTA VALENZUELA (HEXORCISMOS)

​Moisés Horta Valenzuela, known as Hexorcismos, is an artist, technologist, and musician from Tijuana, Mexico, currently based in Berlin. He creates artworks at the intersection of computer music, AI, and digital technology through a decolonial lens. His work has been featured at Ars Electronica, MUTEK, and the Royal Academy of Music in Stockholm. He is the creator of SEMILLA.AI, a neural audio synthesis platform.

​Learn more at http://semilla.ai and hexorcismos.bandcamp.com.

​ABOUT AIxDESIGN

​AIxDESIGN is a non-profit community organization making AI work for the rest of us. We bring people together – like designers, artists, technologists, and researchers – to challenge dominant AI narratives, research alternatives, and make them real. Learn more at aixdesign.co.

​ABOUT OPEN CULTURE TECH 2.0

​This event is hosted by Thunderboom Records + AIxDESIGN as part of Open Culture Tech - an initiative of Thunderboom in collaboration with Netherlands Institute for Sound & Vision, Superposition, AIxDESIGN, Wij Doen Dingen, Bureau Moeilijke Dingen, Bedrijf de Liefde, Chagall, & Wesley Hartogs, and funded by Cultuurloket DigitALL.

​Open Culture Tech makes immersive technology more accessible to artists by running residencies, publishing open-source tools, and hosting showcases to share insights with the music industry. OCT Workshops is an event series hosted by Thunderboom Records and AIxDESIGN during OCT 2.0 offering hands-on workshops featuring musicians using AI and immersive technologies in their creative workflows.