HighCWu

/

FLUX.1-dev-4bit

like 19

Using lora with this model is very interesting. Im going to try this tomorrow, I Am a beginner in this theme, so I will be happy to see any help. Thanks!

Just use adapters exactly as you would with SDXL.

load up all your weights and give them names

    lorapath="whatever_directory/your_lora.safetensors"
    model.load_lora_weights(lorapath,
                            weight_name=lorapath,
                            torch_dtype=torch.float16,
                            local_files_only=True,
                            use_safetensors=True,
                            adapter_name=whatever_you_want_to_name_it,
                            variant="fp16",
                            vae=model.vae,
    )

Then set them as active and set weights

list_of_lora_names=["lora1", "another_lora", "etc"]
list_of_lora_weights = [1.0,.8,.7]

model.set_adapters(list_of_lora_names,adapter_weights=list_of_lora_weights)

You can show what loras are currently loaded:

current_loras = model.get_active_adapters()

and you can delete specific ones:

to_remove=["another_lora","etc"]
 model.delete_adapters(to_remove)