US 11,995,803 B1
Training and deployment of image generation models
Dmitriy Karpman, San Francisco, CA (US); Kevin Guo, San Francisco, CA (US); and Ryan Weber, San Francisco, CA (US)
Assigned to CASTLE GLOBAL, INC., San Francisco, CA (US)
Filed by Castle Global, Inc., San Francisco, CA (US)
Filed on Nov. 30, 2023, as Appl. No. 18/525,628.
Claims priority of provisional application 63/487,552, filed on Feb. 28, 2023.
Int. Cl. G06K 9/00 (2022.01); G06T 5/70 (2024.01)
CPC G06T 5/70 (2024.01) [G06T 2207/20081 (2013.01); G06T 2207/20084 (2013.01)] 20 Claims
OG exemplary drawing
 
1. A method comprising:
receiving a text prompt;
executing a text encoder on the text prompt to generate an embedding representation;
generating a set of base images based on the embedding representation and parameters of a base image generation model;
executing a high resolution model 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;
ranking the set of base images or the set of final images using reward values that are generated by a reward model, wherein 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; and
outputting one or more final images based on the ranking in response to the text prompt.