Generative AI Architecture Pipelines

Tags:
Rapid Prototyping, Model Pressure-Testing, Spatial Logic, Image-to-Image Generation, Facade Ideation

 
The Stack:
ComfyUI, Flux 2 (4B Distilled), Midjourney, Stable Diffusion, Enscape, Rhino, Adobe Photoshop

1. Lobby Custom Stone Render
Feb 2026 (1 day)

1.1 The task: Rapidly visualize a large-scale, custom stone flooring concept within an existing lobby design without spending hours modeling or photoshoping the actual stone pattern.

1.2 The constraints: Must preserve the existing architectural geometry, lighting hierarchy, and furniture placement while achieving photorealistic material replacement that matches the scale and texture of the reference image.

1.3 The inputs:

1.4 The build:
To swap materials without breaking the room’s geometry, I ran a local Flux 2 4B inpaint workflow on ComyUI—prioritizing 4B’s speed to rapidly test ~50 seed and CFG variations. By masking the floor and enforcing a flat, evenly-lit reference texture, I seamlessly integrated the new stone while preserving the furniture and lighting.




1.5 The outputs:


Bad examples:

1.6 After post-production:



2. Facade Design Iteration
Mar 2024 (3 days)