📚 LoRA Loader Stack - Yaser

If you've ever chained five LoraLoader nodes together just to get a face LoRA, a style LoRA, and a detail LoRA all active at once, you know how fast that gets dumb. Each loader is its own node, its own cables, its own little slice of your graph. LoRA Loader Stack - Yaser collapses that whole chain into a single node: one MODEL in, one CLIP in, five LoRA slots, five strength sliders, and one clean output pair at the end.

Why stacking matters

In practice, almost nobody runs a single LoRA anymore. Character work stacks a base-model LoRA with a detail enhancer; style transfer stacks a style LoRA with a quality LoRA; the community's own convention (documented in the LoRA training write-ups in this KB) is that multiple adapters get combined at run time, each with its own strength. The standard way to do that in ComfyUI is a daisy-chain of LoraLoader nodes - and every extra node is a place to accidentally wire the wrong clip or forget a connection when you save and share the workflow. A stack node exists to make the multi-LoRA case one clean, readable box.

The concept is a proven one: this is a direct port of RgthreeLoraLoaderStack from rgthree-comfy, the pack this KB calls "the one you install second, after Manager." If you've used the rgthree Power Lora Loader and loved it, this is the same idea with a fixed five-slot layout instead of dynamic entries.

How it works

Under the hood it's not doing anything exotic - it instantiates ComfyUI's own LoraLoader and calls load_lora five times in sequence, threading the result of each into the next:

• lora_01 through lora_05 - dropdowns populated from whatever .safetensors files are sitting in your models/loras/ folder. Pick "None" to skip a slot.
• strength_01 through strength_05 - per-LoRA weight, from -10 to 10. Negative strengths are a real tool (inverting a style or cancelling a baked-in bias), so don't be shy - though for most LoRAs you'll live between 0.5 and 1.0. The node skips any slot set to "None" or strength 0, so you can leave slots empty without penalty.

The output model and clip are the same types that went in, so this drops into any graph position a normal LoraLoader output would. One cable pair out instead of five.

Install

Same pack, one install gets you everything:

cd ComfyUI/custom_nodes
git clone https://github.com/YaserJaradeh/comfyui-yaser-nodes.git

then restart ComfyUI, or ComfyUI Manager → search "Yaser-nodes". No models to download - it reads your existing LoRA folder. Note the pack's requirements.txt exists but nothing here needs anything beyond stock ComfyUI.

The honest comparison

You've got options. rgthree's Power Lora Loader is more flexible (dynamic list, per-entry toggle, trigger-word lookup) and is the community favorite - if you already have rgthree installed, you may not need this node at all. Where this one wins is simplicity: five fixed slots, no dynamic entries to misplace, and it's already in this pack. If you're installing Yaser-nodes anyway and don't want to pull rgthree just for LoRA stacking, this is perfectly serviceable.

The one gotcha: clip skip and model-only LoRAs aren't handled here. If you need a model-only LoRA (no CLIP effect, common for some character packs) or the trigger-word niceties, that's the rgthree node's turf. For plain "load five LoRAs in order," this does the job with less graph to read.