I'm trying to make a face swap using FLUX+Lora in Python. Currently I stuck using combination of FluxControlNetModel and FluxControlNetInpaintPipeline. While the code works nice using FluxFillPipeline, the results from above combination are really poor. I'm almost sure that I cannot load Lora weights to FluxFillPipeline, so I need to figure out how to use FluxControlNetInpaintPipeline. I tried to use canny and depth with mask, but the results are 100% crap. Could you give me a tip what I do wrong or what should I search? Current full code is available at https://gist.github.com/ptrstpp950/a73da8c7d4cc22b77e8c6c4613b651a6 Continue reading...
