Street-level imagery captured from vehicle-mounted cameras often contains dynamic and undesirable elements such as vehicles, pedestrians, signboards, glare artifacts, or privacy-sensitive regions. These elements limit downstream applications including mapping, urban analysis, visualization, and dataset preparation for computer vision tasks. In this talk, I present a practical, end-to-end inpainting pipeline built using ComfyUI and the Qwen Image-edit diffusion model, and how to make it work for high-resolution street-view imagery.

The workflow combines prompt-guided inpainting with mask-based control to selectively remove or modify objects while preserving structural and semantic consistency. I will walk through how segmentation masks are integrated into ComfyUI graphs, how Qwen Image handles context-aware reconstruction,

I'll also talk about why other inpainting methods failed in this task, and only certain model succeed. Join to learn about how I used Image-edit model, segmentation model, and applied it to high res panoramic imagery