Real-time Neural Rendering

Neural rendering is a technology that could greatly revolutionize the field of computer graphics. By removing the need for textures, materials and rasterization to generate images, we can substantially increase the capabilities of prodecural generation algorithm, since a much lower level of detail is necessary. In place of these, we can use generative AI models to fill in the gaps so that the frames retain a high level of intricacy, and allow us to switch themes, styles, and settings instantaneously. In this project, I will explore the usage of Latent Consistency Models as a method of realtime neural rendering, in conjunction with ControlNet and AnimateDiff.