This ComfyUI workflow leverages the latest model from Flux Tools, black-forest-labs/FLUX.1-Redux-dev, to seamlessly merge images into a cohesive composition. Based on the foundational workflow available at ComfyUI Flux Examples by comfyanonymous, I’ve introduced several enhancements to extend its capabilities.

Unlike the original examples, this workflow utilizes google/siglip-so400m-patch14-384 instead of sigclip_vision_patch14_384 for image processing. Additionally, this workflow merges three images with random weights on each run, enabling you to explore various possibilities. Once you identify the best composition, you can lock in the setup for further refinement. Another key difference is that this workflow concatenates three conditionings rather than connecting them sequentially.

Download Links for Required Models

• Redux Model
• Flux VAE
• Text Encoder (T5 XXL)
• Text Encoder (CLIP-L)
• SigLIP Model

Key Workflow Features

• Disable or Add Input Images: You can disable specific input images or increase the number of inputs by duplicating image node groups.
• Randomized Compositions: Each run generates a new composition with randomized weights for the three input images.
• Node Management: To modify, duplicate, or move nodes, unpin them from the canvas first.

Follow for More Inspiration

To explore more of my AI-generated architectural design concepts, follow my Instagram account at @arch.cantuncok. I share a diverse range of innovative designs brought to life through advanced AI techniques. Join the community and stay inspired by the future of architecture!