Apply Loras

Apply Loras is the hidden workhorse of the Prompt Helpers pack. In the pack's source it's registered under the "hidden" group, and for most people it will never appear on the canvas - because the star of the show, EZ Generate, calls it for you automatically.

Here's the deal. When you write a prompt in this pack's JSON format, you can include lines like:

- lora: my_style
  weight: 0.8

EZ Generate parses that, collects every LoRA in the JSON, and internally expands a chain of Apply Loras nodes - one per LoRA, each becoming a LoraLoader under the hood. You never have to wire a single LoraLoader yourself. That's the entire point of the node existing as a separate class: it's what makes "LoRAs live in my prompt file" work.

When you'd actually touch it

Rarely. The one genuinely useful case is when you want LoRAs applied to a model + CLIP outside the EZ Generate pipeline - say you're keeping your Prompt Helpers JSON as the source of truth for your LoRA stack, but you're sampling with a stock KSampler instead of EZ Generate. Then Apply Loras is your bridge: feed it the MODEL and CLIP from an EZ Checkpoint, hand it the LoRA list, and route its MODEL/CLIP outputs into whatever sampler you like.

Inputs and outputs

Three inputs, two outputs, no knobs to fiddle with:

• model (MODEL) - the model to apply LoRAs onto.
• clip (CLIP) - the text encoder, also gets the LoRA.
• loras (LORAS) - the list of LoRA paths and weights.

Outputs are MODEL and CLIP, ready to feed onward. The loras input is a custom type the pack itself produces - you won't be hand-feeding it from a stock node; it's built to receive the pack's internal list.

How it resolves your LoRA files

The lookup has a few sharp edges worth knowing:

• If a path has no extension, it appends .safetensors for you.
• It resolves against ComfyUI/models/loras - so drop your files there, not anywhere else.
• Missing file? You get Could not find lora: <path>, and the run dies loudly rather than silently producing an image with no LoRA.
• The same LoRA applied twice in one JSON is a hard error ("Duplicate lora"), which is this pack being stricter than you're used to - stock ComfyUI happily applies a LoRA twice and compounds it.

Installing it

Install the whole pack once - Apply Loras ships with it. In ComfyUI Manager, search "Prompt Helpers" (the repo is Pauan/comfyui-prompt-helpers) and install. Or, manually:

cd ComfyUI/custom_nodes
git clone https://github.com/Pauan/comfyui-prompt-helpers

Then restart ComfyUI. No model downloads, no heavy dependencies - the pack only needs PyYAML and the desktop-notifier library (for the EZ Notify node), and it requires ComfyUI 0.8.0+ since it's built on the newer Nodes 2.0 API.

The most common gotcha is the one above: a LoRA filename that doesn't match, or a file sitting in the wrong folder. Double-check the filename in your JSON against models/loras - the error message will name the exact path it looked for, so read it before re-typing.