Materia Mania is a collection of abstract, AI generated forms set against a solid black background. This project highlights the emergent and chaotic nature of a diffusion model, that was controlled to create simple, elegant shapes and textures. The resulting "materia" is intentionally representative and aesthetically pleasing to showcase the possibilities of combining human creativity with technological advancement. The collection is a unique example of the transformative power of AI art and its ability to reshape the future of the medium.
Materia Mania was created using an iterative process involving generative coding, aesthetics modeling, and AI augmentation. The foundation of each image was defined by simple combinations of geometric shapes and colors on a solid background. A custom generative-art diffusion model was fine-tuned on 120K image-text pairs to produce high-fidelity abstract augmentations of these base images. 25K base images were initially generated and incrementally upscaled to a resolution of 4096x4096. An aesthetics rating system and model were developed to select the most aesthetic outputs. The resulting collection showcases the unique and visually striking potential of combining state-of-the-art generative models with custom fine-tuning and aesthetics modeling.
Huemin is a digital AI artist and programmer who discovered a passion for AI art as a medium in 2021. Since then, Huemin has been creating digital artworks that showcase a unique blend of color, minimalism, and geometric abstraction.
As an innovator in the AI art space, Huemin has not only created a notable body of work but has made significant contributions to the open-source community. Huemin has developed AI art tools that have been widely used by other artists such as JAX Automated Stitching and led community projects such as Deforum Stable Diffusion, a free to use stable diffusion inference tool.
Huemin also mantains the deforum community discord which has over 15K members and is a breeding ground for cutting edge AI implementaions.
