Diffusion Pipelines for Urban Scene Editing

What is it?

This project provides a set of open-source, fully local workflows for text-guided transformation of architectural and urban scenes. Built on ComfyUI and diffusion-based image editing models (currently Qwen-Image-Edit in GGUF format), the pipelines allow natural-language manipulation of urban imagery — without any reliance on cloud APIs, external GPUs, or proprietary services.

Users can modify streets, façades, and public spaces through text prompts: adding vegetation, changing building materials, converting roads to pedestrian plazas, or compositing elements from reference images. Everything runs offline, ensuring data privacy, reproducibility, and fine-grained control over every transformation.

The project is distributed as a collection of versioned ComfyUI workflow files plus custom Python nodes developed by Urban Geo Analytics. Each version introduces new capabilities — from basic single-image editing to reference-based composition, sequential batch processing, and mask-guided inpainting.

Versions & Capabilities

Step v1.0: Single-image editing. Basic text-guided editing with automatic aspect ratio adaptation. Prompt examples: “Add trees along the sidewalk”, “Turn this street into a pedestrian plaza.”

Step v1.1: Reference-based editing. Edit scenes using a second image as material or style reference. Example: “Change the walls to the brick material from image 2.”

Step v1.2: Sequential and batch edits. Custom Python nodes for automated iteration over image folders. Apply the same transformation to hundreds of urban photographs.

Step v1.3: Mask-guided inpainting. Localised editing with spatial control. Paint a mask region and composite elements from a reference image into that area. Example: “Add the cyclist from image 2 into the mask region.”

Illustrations

Key Features

• Text-driven urban editing. Modify streets, façades, vegetation, and public spaces through natural language prompts.
• Reference-based composition. Use a second image as a material, style, or object source for context-aware transformations.
• Mask-guided inpainting. Spatially targeted edits: define where in the scene to apply the transformation.
• Batch processing. Custom Python nodes for sequential iteration over image folders. Scale edits to entire datasets.
• Fully local and private. Runs entirely on your machine. No cloud APIs, no remote inference, no data leaves your system.
• ComfyUI integration. Modular node-based workflows. Import, customise, and extend through the ComfyUI interface.

Applications

• Urban design prototyping. Rapid visualisation of streetscape interventions — greening, pedestrianisation, façade renovation — before any physical change.
• Participatory planning. Generate visual scenarios for public consultation. Show residents what proposed changes would look like on their street.
• Architectural visualisation. Material substitution, vegetation placement, and context-aware scene composition for design presentations.
• Dataset augmentation. Generate training data for downstream computer vision models by systematically transforming existing urban imagery.

Download & Installation

1. Download and install the git version of ComfyUI with ComfyUI Manager, or the standalone ComfyUI (already includes the Manager).
2. Open the desired workflow and when prompted by the manager install the missing nodes. Then, follow the Qwen-Edit-UGA tutorials to download and place the necessary checkpoints, VAE, CLIP, and other model files inside your ComfyUI/models directory.
3. To directly add our custom nodes (required from v1.2), navigate to your ComfyUI directory: ComfyUI/custom_nodes/ and clone or download our repository and put it in your custom_nodes directory
   git clone https://github.com/perezjoan/ComfyUI-QwenEdit-Urbanism-by-UGA.git
4. Restart ComfyUI

The nodes will appear under image/sequence and image/random categories.

Tutorials & Workflow Downloads

Open ComfyUI → File → Load → Workflow and choose the version file to import it into your workspace.

Credits

Developed by Urban Geo Analytics (UGA)
Based on open-source work by QuantStack, ComfyUI, and Qwen Image Edit contributors.

🔗 GitHub Repository: github.com/perezjoan/ComfyUI-QwenEdit-Urbanism-by-UGA

🪪 License: MIT
© 2025 Urban Geo Analytics — All rights reserved.