[BUG] - Model output impairs LoRA application

[alessandroperilli]

Description

First: it took me countless hours over an entire week to pinpoint this behavior. Yet, I am not sure this is a bug. It might be that I don't understand well enough the intricacies of ComfyUI. With that in mind, here's the problem:

When the Model output of an SD Parameter Generator node is passed to an Efficient KSampler node, whatever LoRA has been defined in a Efficient Loader node gets ignored.

Reproduction steps

Recreate the following situation:

As you can see in this example, the Cartoon_SDXL_V2 LoRA is completely ignored by the Efficient KSampler node if I use the Model output generated by the SD Parameter Generator node.

However, if I use the Model output coming straight from an Efficient Loader node, the LoRA is correctly applied:

In both cases, notice this is an img2img workflow simply VAE Encoding a source image.

If this is not a bug, but the expected behaviour, it makes impossible for me to use this portion of the SD Parameter Generator node in the AP Workflow, so I really hope it's a bug.

Thank you!

Image file

Output when the Model comes from SD Parameter Generator:

Output when the Model comes from Efficient Loader:

[receyuki]

I've set up a new workflow for testing, and everything is working fine, including Lora. However, when I tried to understand your workflow, I noticed that you forgot to apply Lora to the model. You need to apply Lora to both the model and the clip for it to work properly

[alessandroperilli]

I realized this morning, but didn't have a second to close the issue. Sorry about that.

Efficiency nodes are amazing, but sometimes they make you forget how many things they do behind the scene and if you change something, you make these silly mistakes. Sorry again.