The Birth of Vidia

In late 2022, a Corridor Digital video sparked a rabbit hole that, as a film production student at USC's School of Cinematic Arts, seemed irresistible. The spark came from seeing how Stable Diffusion could transform live-action footage into anime. It got me thinking: what if we could reverse this process? Could we use AI to turn basic CG into photorealistic imagery? This idea coincided with my growing interest in virtual production, setting the stage for an ambitious thesis project.

I aimed to recreate the final "Ex Machina" scene using Mocap and Unity. However, the initial results left much to be desired. Determined to achieve something groundbreaking, I dove headfirst into AI-enhanced rendering. What followed was a period of intense learning and experimentation. The landscape of generative AI was changing rapidly, with new developments emerging daily. As someone without a formal computer science background, I grappled with complex concepts like epochs, stochastic gradient descent, and the intricacies of training neural networks. The challenges were numerous. Resources were scarce, and the information available was often cryptic or contradictory. I had to navigate unfamiliar territories, from renting GPU power to fine-tuning hyperparameters. There were days when a new release would render weeks of work obsolete, forcing me to adapt quickly and constantly.

Despite these hurdles, I pressed on, driven by the potential I saw in this technology. My persistence paid off in unexpected ways: The AI didn't just mimic the movie's style; it demonstrated an uncanny ability to predict physical properties I had never explicitly taught. It accurately rendered subsurface scattering, simulated realistic lighting, and even got muscle movements right. These emergent capabilities were as surprising as they were exciting.

Presenting my thesis to my professors and peers was a mixed experience. While there was fascination, there was also skepticism. Many viewed AI as a shortcut, failing to grasp the complexity and effort involved in achieving these results. I continued to refine my process and transitioned from Auto1111 to ComfyUI, creating an extensive workflow that built upon my earlier experiments. This culminated in a new project where I seamlessly blended stylized and photorealistic characters in a single video.

Throughout this process, a realization began to dawn on me. The power of this technology was immense, but its complexity made it inaccessible to many content creators. The steep learning curve and technical barriers were preventing others from harnessing its full potential. This realization was the genesis of Vidia. I saw an opportunity to democratize this technology, to create a tool that could make AI-powered rendering accessible to all content creators. Vidia isn't just about enhancing CG or improving VFX; it's about empowering creators to bring their visions to life, regardless of their technical expertise.

With Vidia, filmmakers and artists can focus on their creative vision, using the tool to elevate their work to new heights of realism and style. Whether it's improving lighting, filling in details, or completely transforming the look of footage, Vidia opens up a world of possibilities that were previously out of reach for many creators.

Looking back on this journey, from those first experiments with Stable Diffusion to the development of Vidia, I'm struck by how far the technology has come—and how far it still has to go. The road ahead is exciting and full of possibilities. As AI continues to evolve, so too will Vidia. I hope it will empower a new generation of creators, enabling them to bring their wildest imaginings to life with unprecedented ease and realism.

The future of filmmaking is here, and it's more accessible than ever.