Autolume - A No-Code AI Visual Synthesizer

"Autolume is a no-code visual synthesizer developed by the Metacreation Lab. It enables artists to train and explore their own models with small datasets, giving them direct creative control and the ability to perform live with AI-generated imagery. The system covers the full workflow, from data preprocessing and model training to real-time latent space navigation and output upscaling. By making the artistic potential of generative AI accessible to non-technical users, Autolume supports a hands-on workflow that fosters creative ownership."

"It also integrates with the OSC (Open Sound Control) protocol for audio-reactive visuals, making it a powerful tool for live performance. Autolume is built on Generative Adversarial Networks (GANs) and provides a controllable artistic workflow. The Autolume-live module allows real-time latent space exploration and network parameter control via OSC for audio-reactive works and other interactive works. Built on GANs, Autolume supports interactive applications through OSC integration."

"Model Training: Train models from scratch or resume from a checkpoint, with support for square and non-square datasets. Augmentation techniques enable training with small datasets. Latent Space Projection: Project an image or text embedding into the latent space for unique artistic exploration. Model Mixing: Blend two trained models into a new one, combining their visual features. Super-resolution: Upscale images and videos with a dedicated module for high-resolution output."

Autolume is a no-code visual synthesizer that lets artists train and explore GAN models using small datasets and augmentation. The system manages the entire workflow from data preprocessing and model training to latent space projection, model mixing, and super-resolution upscaling. Autolume supports square and non-square datasets, training from scratch or resuming from checkpoints. The Autolume-live module enables real-time latent space navigation and network parameter control via OSC for audio-reactive and interactive performances. Latent space projection accepts images or text embeddings for targeted exploration. Model mixing blends two trained models into new visual styles.

Read at CreativeApplications.Net