MUMBAI, India, Sept. 22 -- Intellectual Property India has published a patent application (202611091007 A) filed by Manipal University Jaipur on July 27, 2026, for Low-Compute Diffusion Model For Image Generation In Mobile Devices.
Inventors include Mansi Negi; Divyansh Mishra; Subhash Singh; Simran Kaur; and Siddhanta Kumar Singh.
The application for the patent was published on September 18, 2026, under issue no. 38/2026.
Abstract: The present invention relates to a low-compute diffusion-based image generation system and method for generating synthetic images on mobile devices and resource- constrained computing platforms. The system comprises a noise generation module configured to initialize an image generation process using Gaussian noise, a diffusion scheduler configured to perform a reduced number of diffusion steps, and a lightweight U-Net architecture comprising encoder-decoder blocks, residual blocks, time embedding layers, and linear attention mechanisms for efficient denoising operations. The system further comprises a DDIM-based sampling module configured to perform step-skipping reverse diffusion to reduce inference time and computational complexity, and an image reconstruction module configured to generate output images from denoised latent representations. The invention utilizes a cosine noise scheduling mechanism and a reduced diffusion schedule to achieve efficient image generation while maintaining acceptable image quality. The proposed system is capable of operating on mobile devices and edge computing platforms using central processing unit resources without dependence on dedicated graphical processing units or cloud infrastructure. The invention significantly reduces memory consumption, energy requirements, and image generation time, thereby enabling practical deployment of diffusion-based generative artificial intelligence systems in resource-constrained environments.
Disclaimer: Curated by HT Syndication.
