I've worked on this flow for a week, striving to make it as beautiful as possible. When I began experimenting with SDXL 0.9, returning to SD 1.5 was not an option;

it's same of the transition from using GPT-3 to GPT-4, where reverting back to GPT-3 becomes challenging :D

Therefore, I endeavored to create a flow that can manage various types of prompts and styles, including the integration of a Lora stack loader.

Simply adjust the latent size, add your preferred Loras and model, and let your imagination take over.

Update:

I found issue with the clip vision nodes, where the prompt has the higher hand over the generated image, so i replaced it with Depth controlnet node, you can adjust the strength of the reference image , in Apply Controlnet node.