Lora Stack Loader 📚

Once a workflow needs more than two LoRAs, the classic chain - LoraLoader → LoraLoader → LoraLoader, each one eating the previous model output - turns into a horizontal snake that takes over your canvas and makes every tweak a chore. IkkiLoraStackLoader is the pack's answer: one node with a built-in list UI where each row is a LoRA with its own enable toggle and strength, and the whole stack applies in a single pass. It's the same problem rgthree's Power Lora Loader solves, in a smaller, self-contained form.

The strength is in the widget. Click "+ Add LoRA" to add a row; each row has a green/red toggle to enable or disable that LoRA without deleting it, a dropdown listing every .safetensors in your models/loras folder, and a strength box you drag horizontally to scrub the value. That drag-to-scrub is the genuinely nice touch - you can sweep strength from 0.1 to 1.2 live and watch what happens downstream.

How it works

Technically, the list is a hidden JSON string field that a custom frontend widget fills in. The Python side parses it, and for each enabled entry it loads the LoRA weights and applies them via ComfyUI's own load_lora_for_models - the exact same code a normal LoraLoader runs. So you're not getting a different result, just a cleaner graph. Rows that are disabled, or point at a missing file, are skipped (and logged) rather than crashing the whole stack.

Inputs and outputs

• model and clip - feed the output of your checkpoint loader into both, exactly like a LoraLoader.
• lora_stack - the hidden JSON field the UI maintains. Don't hand-type this; the widget writes it.
• Outputs MODEL and CLIP, ready for your sampler and text encoder.

One strength applies to both model and CLIP on each row (there's no separate model/CLIP strength split, unlike some loaders) - for most LoRAs that's what you want.

Install

Via ComfyUI Manager (search "ikki") or:

cd ComfyUI/custom_nodes
git clone https://github.com/RedsAnalysis/comfyui-ikki-pack
# restart ComfyUI

No extra pip dependencies - it's pure ComfyUI internals.

Where it falls short

• No per-LoRA model vs CLIP strength. Power Lora Loader gives you both dials; this gives you one. For character LoRAs that's fine; for style LoRAs where you often want to back the CLIP strength off, you can't.
• No trigger word display. rgthree's loader reads trigger words off the file or from CivitAI; this one just lists filenames.
• Disabled rows are skipped, not zero-strength. If you need a LoRA present but neutral for a comparison, you can't express "strength 0" - you'd disable it, which isn't identical in every case.

Honest ranking: if you already run rgthree, Power Lora Loader is richer and you don't need this. But it's a clean, dependency-free alternative for people who want the Ikki pack's own UI, and the drag-to-scrub strength is something Power Lora doesn't do as directly. For a young pack it's one of the more polished nodes - it's one of the few the README actually documents, and the pack's own example workflow (an Anima upscaler) uses it, so you know the intended pattern works.