A workflow to explore the flux1.s output as a latent image, and use the SDXL to refine it, upscale it, and detail it to the final image.
On the KSampler (Efficient), feel free to use as many steps as you want. I saved the workflow with the settings that worked best for me.
I made it for personal use, but decided to share it on a whim xD.
My personal recommendation is to use both Serget's model: ICH_WILLl_MEIN_STEIL v4
Alexsd9's model: Babes XL 3.0
But it should work with SD 1.5, SDXL, and a friend said it works with Pony too, but I didn't test it.
This workflow was made for a PC that only have 32GB system ram and 12GB of vram.
If you need small ControlNet models: Go here!
Also, I recommend using Mistolines if you want to increase the accuracy between Flux output and SDXL output.
Hope it works for you as well as it did for me!
Description
Added ControlNet, best used with small controlnet-xl-models.
Added Lora Stacker for Flux Loras.
FAQ
Comments (2)
I am getting this with the Ksampler (efficient) node:
Error occurred when executing KSampler (Efficient): 'TimestepEmbedSequential' object has no attribute '1'
Check if you're using the correct controlnet models, if you're SD checkpoint is SDXL, you should use an SDXL controlnet model, and subsequently if using a SD 1.5, I left the links for the controlnet models in the description.