Empowering Artistic Control in GenAI

Building on six years of generative AI experimentation, BeautyRender showcases the potenticial on advancing artistic control in generative settings. The project's focus was on allowing highly realistic and controllable lighting, posing, and texture details, particularly in skin and fabric.

BeautyRender is a culmination of five months of intense research and development, a Stable Diffusion checkpoint created with my own custom dataset of high-resolution textures, including skin, hair, fabric, and various materials. The model offers great control over race, age, and gender, ensuring highly accurate and diverse representations.

It allows easy control of light, shadow, and focal distance, along with surface textures (wet or dry) for enhanced realism. BeautyRender ensures character consistency across different poses and expressions, thanks to an intuitive gesture-based multi-touch UI for head poses and facial control: touch-and-drag to pose, and pinch to manipulate expressions.

This project represents a relevant achievement in my generative AI experiments, allowing a great level of realism and control in image generation. The successful integration of advanced texturing and UI design confirms the potential for much better creative AI workflows.