It's the evolved version of my first 1.5 model. This Lora is for experimental purposes, it's trained on my architectural study images to stir up the architectural discipline and give some critical ideas. I used khoya to train it.

Kubakub_flux1_architecture

This time I added more diversity to the dataset, including over 700 drawings, renders and photographs.I would start with a strength of 0.6-0.75. I started also experimenting with the quantized version, Q8 does a good job with them, but still, the original dev gives more interesting results.

Files with SHA256: 9cf90460e9aefab103fe7f0df340f5fd9cc54d3432a275ee44c55991c90b5635