>
I
Introduction
Description: Imagen 4 is a latent diffusion model that generates high quality images from text prompts. Imagen 4 performs well in photorealistic composition seings and has improved spelling and typography, instruction following and richer colors, textures and details compared to previous Imagen models. Inputs: The inputs consist of natural-language text strings (e.g. instructions for creating a synthetic image using a visual description) or image files. Outputs: Outputs are generated high quality images in response to text and image inputs. Architecture: Imagen 4 utilises latent diffusion, which is the de facto standard approach for modern image and video models, achieving high quality performance in generative media applications.
Authors
Gabriel Barcik, Jakob Bauer, Dana Berman, Nicole Brichtova, Lluis Castrejon, Matan Cohen, Sander Dieleman, Yuqing Du, Praneet Dutta, Jess Gallegos, Yilin Gao, Evgeny Gladchenko, Susan Hao, Ruba Haroun, Ed Hirst, Tobenna Peter Igwe, Xuhui Jia, Siavash Khodadadeh, Pavel Khrushkov, Karol Langner, Rory Lawton, Yinxiao Li, Yandong Li, Shixin Luo, Michael Mathieu, Soňa Mokrá, Aäron van den Oord, Lily Pagan, Zarana Parekh, Noam Petrank, Jordi Pont-Tuset, Hang Qi, Deepak Ramachandran, Poorva Rane, Ali Razavi, Robert Riachi, Dirk Robinson, James Thornton, Felix Riedel, Evgeny Sluzhaev, Hansa Srinivasan, Srivatsan Srinivasan, Benigno Uria, Cristina Vasconcelos, Oliver Wang, Simon Wang, Austin Waters, Daniel Winter, Chris Wolff, Xin Yuan, Zhisheng Xiao, Keyang Xu, Andrew Xue, Katie Zhang, Yang Zhao
Language
Multimodal
Vision
Earth science