Patent · US Active

Training and deployment of image generation models

US11995803B1 · kind B1 · utility

4Cited by
0References
20Claims
0Family size

Assignee

Inventors

Key dates

Filing dateNov 30, 2023
Grant dateMay 28, 2024
Priority date
Expiry dateNov 30, 2043

Classification

  • Technology area (CPC G)Physics
  • CPC primaryG06T2207/20084
  • WIPO fieldComputer technology
  • WIPO sectorElectrical engineering

Abstract

In some embodiments, a method receives a text prompt and executes a text encoder on the text prompt to generate an embedding representation. A set of base images is generated based on the embedding representation and parameters of a base image generation model. A high resolution model is executed to upsample one or more base images in the set of base images based on parameters of the high resolution model to generate a set of final images. The method ranks the set of base images or the set of final images using reward values that are generated by a reward model. The reward model is trained using human input that provided feedback on a quality of generated images using the base image generation model and the high resolution model. One or more final images are output based on the ranking in response to the text prompt.

Source: USPTO / EPO open patent data. Objective bibliographic and citation counts.