Art / Low CFG
Low CFG images are generated with minimal classifier-free guidance, letting the diffusion model express its own learned visual logic rather than following a text prompt. The results are abstract, strange, and deeply structured — revealing the latent patterns compressed from billions of training images.
The process involves tuning model parameters, sampling strategies, and generation settings to coax out certain qualities — then curating from hundreds of outputs to surface the images that resonate. It's part science, part intuition, and part visual archaeology.
