I have been developing a custom workflow in ComfyUI in collaboration with a designer, focused on accelerating concept iteration and improving visual consistency during early-stage design.
The pipeline leverages a series of purpose-trained LoRAs across both Flux Dev and Kontext, each fulfilling a distinct role:
- A Flux LoRA, trained on a targeted dataset of specific vehicle types and configurations, is used to generate the initial concept imagery.
- Multiple Kontext LoRAs are then applied to refine secondary views (side and rear), ensuring greater structural accuracy and coherence across perspectives.
- A final Kontext LoRA transforms the outputs into sketches in the style of the designer, allowing for flexible adaptation to different visual languages.
The initial implementation focuses on a clean, minimal sketch aesthetic. Future iterations will explore more expressive and gestural styles to broaden the creative range of the system.
This workflow demonstrates how generative tools can effectively bridge the gap between concept and visualisation — enabling faster iteration while maintaining a high degree of artistic control.