Trained Lora makes no changes to images

Custom Nodes
1

Hello, I’m trying to train my Lora, I’m using one of these Lora training templates flux_lora_train_example.

I have bunch of photos in certain style, so I skipped blip captioning. left trigger word empty, put sample prompts and mostly didn’t changed any settings.

I could actually see in validation images that training output really follows the style from input images.

But after training I’m taking that file from flux_lora_output_path folder, putting it in Lora folder, restarting ComfyUI, opening lora template and then that lora makes literally zero changes to my image, even if i set strength to 100.

I’ve tried quickly train other models with changing stuff and setting less training steps, but still, all my Loras do nothing

Can anyone tell me what am I missing?

obraz
obraz1920×1646 143 KB

2

I’m not sure the exact reason so please consider the following.

Make sure you are using the keyword for the LoRA in the prompt.
Before I train my LoRA in OneTrainer, I run all my images through ComfyUI node “Lora Training in ComfyUI” which generates text files with tags. The first tag is always the name of your LoRA. For example, if I name it “Naruto1”, I can then put “Naruto1’“ in my positive prompt to activate the LoRA.

Furthermore, if your cfg is too high or denoise too low, it will prevent the image from changing enough to match the LoRA training images.

I can’t see all of your settings, but it does look good in terms of node linkage.
Please try these and let us know.

1 Like