Why Image Generation Needs More Than Bigger Models with Fatih Porikli

Text-to-image models have become remarkably good at producing realistic images, but Fatih Porikli argues that realism is not the same as correctness — models still fail at several distinct people, precise compositions, and local high-resolution generation. The Qualcomm Vice President of Technology walks through approaches his team presented at CVPR: better training objectives for controllability, separating scene planning from rendering, and generating 16-megapixel images efficiently on edge devices. The conversation also covers eliminating visible artifacts in AI-powered editing, reinforcement learning for image generation, and agentic pipelines. Porikli maps what remains unsolved in generative vision and where the next phase of progress is likely to come from.