This AI Is Reshaping How We Build Game Worlds

In this video, we cover 2 unique Machine Learning principles- Discrete Absorbing Diffusion, and Vector Quantized Variational Autoencoders- and see how they can be applied to generate new, highly controllable worlds in Minecraft. This work is currently still in progress, and updates including the final paper and additional demos will be added over time This video is in collaboration with NYU’s Game Innovation Lab and covers work by Tim Merino and collaborators. Tim Merino’s Personal Site: https://timmerino1710.github.io/ NYU Game Innovation Lab Site: https://game.engineering.nyu.edu/ Additional References and Resources: Written Explanation for Discrete Absorbing Diffusion Models: https://www.ml6.eu/blogpost/using-generative-ai-for-image-manipulation-discrete-absorbing-diffusion-models-explained VQGAN Paper/Explanation: https://arxiv.org/abs/2012.09841 Potential Applications of Discrete Diffusion Models: https://www.ml6.eu/blogpost/potential-applications-of-discrete-diffusion-models Unleashing Transformers: Parallel Token Prediction with Discrete Absorbing Diffusion for Fast High-Resolution Image Generation from Vector-Quantized Codes: https://samb-t.github.io/unleashing-transformers