USPTO Patent Grant: Non-adversarial image generation using transfer learning

ChangeBridge: Patent Grants - AI & Computing (G06N)

Published March 24th, 2026

Detected March 25th, 2026

Summary

The USPTO has granted Adobe Inc. a patent (US12586258B2) for a method of non-adversarial image generation using transfer learning. The patent describes a system that generates latent representations and digital images based on random noise and a trained machine learning model, aiming for visual similarity to a training dataset.

View original document View source feed page

What changed

The United States Patent and Trademark Office (USPTO) has issued patent US12586258B2 to Adobe Inc. This patent covers systems and methods for non-adversarial image generation utilizing transfer learning. The core innovation involves a generation system that takes random noise as input, processes it through a transformer model to create a latent representation, and then uses a machine learning model to generate a digital image that is visually similar to a training dataset, based on a perceptual loss.

This patent grant is primarily an intellectual property matter and does not impose new regulatory obligations on businesses. However, it signifies a development in AI image generation technology, which may be relevant for companies operating in the AI, software, and digital media sectors. Companies developing or utilizing similar AI image generation techniques should be aware of this granted patent and its potential implications for their intellectual property landscape.

Source document (simplified)

← USPTO Patent Grants

Non-adversarial image generation using transfer learning

Grant US12586258B2 Kind: B2 Mar 24, 2026

Assignee

Adobe Inc.

Inventors

Puneet Mangla, Balaji Krishnamurthy

Abstract

In implementations of systems for non-adversarial image generation using transfer learning, a computing device implements a generation system to receive input data describing random noise. The generation system generates a latent representation in a latent space of a machine learning model based on the random noise using a transformer model that is trained to generate latent representations in the latent space. A digital image is generated using the machine learning model based on the latent representation that depicts an object that is visually similar to objects depicted in digital images of a training dataset used to train the machine learning model based on a perceptual loss.