Karan Sood

EXCELLENCE PERFORMANCE SAMSUNG Nov 2024 At Samsung PRISM, I worked on developing subject-driven GANs and conditional GANs (cGANs) inspired by DreamBooth to enhance personalized image generation. I significantly improved image quality by increasing resolution and integrating a text-to-image pipeline using LLaMA and Stable Diffusion for more accurate prompt-to-image outputs.I also fine-tuned Stable Diffusion using LoRA on a focused 50-image product dataset, optimizing the img-to-img pipeline for low GPU-RAM environments. The entire… Show more At Samsung PRISM, I worked on developing subject-driven GANs and conditional GANs (cGANs) inspired by DreamBooth to enhance personalized image generation. I significantly improved image quality by increasing resolution and integrating a text-to-image pipeline using LLaMA and Stable Diffusion for more accurate prompt-to-image outputs.I also fine-tuned Stable Diffusion using LoRA on a focused 50-image product dataset, optimizing the img-to-img pipeline for low GPU-RAM environments. The entire pipeline was lightweight, efficient, and capable of generating high-fidelity, customized images. Show less

HACKHO JAVA HACKHATHON HACKHOS BY NEWTON SCHOOL May 2024 3 position