In this case, it seems that FLUX inference is mostly done in-house. Therefore, it seems quite difficult to apply LoRA at runtime using conventional methods such as Diffusers.
If you upload the entire model with LoRA applied to Hugging Face and replace this file to load the applied model, it should work.