Imagic: Text-Based Real Image Editing with Diffusion Models
Demonstrates, for the very first time, the ability to apply complex (e.g., non-rigid) text-guided semantic edits to a single real image using Imagen.
Offers businesses the ability to apply complex, text-guided semantic edits to real images in a single, unified framework, reducing the need for multiple input images and additional inputs, such as image masks or additional views of the object.
Challenging BIG-Bench Tasks and Whether Chain-of-Thought Can Solve Them
PaLM / Codex + CoT outperforms the average humanrater performance on many of the 23 challenging BIG-Bench tasks.
Provides insights into how language models perform on challenging tasks, and how chain-of-thought (CoT) prompting can improve model performance, thereby offering businesses improved natural language processing capabilities.
LAION-5B: An open large-scale dataset for training next generation image-text models
The arXiv paper for LAION-5B. Shows successful replication of foundational models like CLIP, GLIDE and Stable Diffusion.
Offers businesses access to a large-scale dataset consisting of billions of CLIP-filtered image-text pairs, enabling improved training and capabilities of language-vision models, without requiring expensive and accurate labels used in standard vision unimodal supervised learning.