machine learning

Mums Against Donk 4

Machine learning and real-time generative visuals for London's leading donk and happy hardcore club night.

Building on my existing work with the collective, I created a visual set that pays homage to their event's reputation as being camp and silly. With TouchDesigner, I designed a set of generative components that respond in real-time to the incoming audio from the decks. These components are inspired by liquid light shows, and make use of generative growth systems, noise displacements and video distortion.

To complement Alterum's DJ set, I used transfer learning to manipulate a GAN model to create abstracted, morphing looping visuals that were displayed behind her throughout. Alterum always wears distinctive makeup during each of her sets, and this has become synonymous with Mums Against Donk's visual identity.

by Alicia Wright

INFERNO Visuals Dec 2023

Audio-reactive generative visual set controlled, rendered and delivered in real time using TouchDesigner.

For INFERNO's final event of 2023, I applied creative machine learning, GLSL, 2D textures, deformed 3D meshes and manipulated found footage to create a visual environment that is an abstracted representation of INFERNO's history and community.

The visual set consisted of generative components that respond in real time to a live audio feed from the decks, via the venue's main sound desk. The components were programmed to respond to low frequencies, moving in time to kick drums and heavy bass as a result, with additional custom parameters controlled on the fly using MIDI.

I trained a GAN using over 2,300 images pulled from the INFERNO photo archive's almost 9-year history. The model was trained under StyleGAN 2 ADA at 512px for 550kimg. Latent vector interpolation loops from randomly selected seeds featured heavily throughout the set, and incoming low audio frequencies were used to control their playback.

Also included in the visual set were found video clips of lava flows and volcanic eruptions, digitally manipulated to move in time with the live music, and creatively recoloured.

by Alicia Wright

Mums Against Goffs Machine Learning Visuals

Generative live visuals for London's leading donk and happy hardcore club night's crossover event with r u a goff?

In collaboration with photographers El Hogg @999999999boyscrysendpics & Mia Evans @ahgeewiz, I produced a collection of generative live visuals as part of a larger visual set shown in the event's main room.

Using Nvidia StyleGAN2 ADA PyTorch, I trained machine learning models from scratch using datasets of images taken at previous Mums Against Donk events. The variety within the datasets, and the reduced training time allowed me to create highly abstracted motion graphics which interpolate between different latent vectors.

The final datasets consisted of ~280 images each at 256px. El's model was trained for 550kimg and Mia's for 750kimg.

The resulting works are a unique representation of the community cultured by Mums Against Donk’s parties, paying homage to the ever morphing identity of London's queer nightlife.

by Alicia Wright