OpenAI, one of the most influential research institutions in the AI field, is investigating a new publication-worthy technique called “consistency models”. Diffusion models, the most popular method of generating images seen today, generate their results with lengthy and computationally intensive processes. Whereas the new consistency models by OpenAI can be used to generate images orders of magnitude faster and with much fewer iterations.
This technical research paper focuses on the advances made in image generation and execution, as OpenAI believes that consistency models are the “next big thing” when it comes to AI imagery. According to their research, these models need very few computations, and as fast as one or two steps, they can create realistic images by processing gradually decreasing noise from original noise-only images.
Whereas the more time consuming diffusion models generally have to go through 10 to 1000 different iterations to complete their task. With faster images produced through consistency models, applications like live chat interfaces and phone applications which rely on image generators will be able to use AI imagery without draining their batteries.
OpenAI researchers like Ilya Sutskever, Yang Song, Prafulla Dhariwal, and Mark Chen are actively searching for the best way to leverage these models in tasks like colorizing, sketch interpretation and upscaling. In addition, these capabilities can be improved and enhanced with a short second step of computation.
Founded in 2015, OpenAI is a San Francisco-based artificial intelligence research lab focused on developing natural language processing (NLP) and reinforcement learning (RL) systems. The team at OpenAI consists of world-renowned researches and scientists such as Ilya Sutskever and Greg Brockman, who are looking to shape the future of artificial intelligence and machine learning technology.
Ilya Sutskever, the co-founder of OpenAI, is a graduate of the Hebrew University in Jerusalem, where he completed a Bachelor’s and Master’s degree in mathematics. He went on to complete a Ph.D in computer science from the University of Toronto, making him a world leader in the field of artificial intelligence and machine learning. He is actively researching techniques such as consistency models with his work at OpenAI, breaking boundaries and redefining the possibilities of AI.
