We are exploring how to build a structured diffusion pipeline for architectural visualization.
In our case we often deal with full residential developments with multiple housing types rather than a single curated building. That means the system needs to generate many images while maintaining consistency and reproducibility across outputs.
We are experimenting with workflows using Stable Diffusion / SDXL, ControlNet and ComfyUI.
The pipeline will likely need to work with structured inputs such as:
• depth maps
• segmentation maps
• reference imagery
• possibly 3D model data
Curious if anyone here has worked on diffusion pipelines for architectural or large scale visual generation.
If this sounds interesting or similar to work you’ve done, feel free to reply or send a message.