Load Lora Stacked Advanced with Previews (XTNodes)

Stack several local LoRAs onto one chain, mute any of them without unplugging a wire, and - the "Advanced" part - tune how hard each one pulls on the model separately from how hard it pulls on the clip/text encoder. That last bit is the whole reason this node exists over its plainer sibling, LoraLoaderStackedWithPreviews, which forces the same number onto both.

How it works

This node doesn't apply anything itself - it builds a LORA_STACK, a plain list of (file, model weight, clip weight) entries, that you feed into an "Apply LoRA Stack" node downstream. The pack's README explicitly calls out that its stack format is interchangeable with Comfyroll's ComfyUI_Comfyroll_CustomNodes, so you can mix this pack's stackers with Comfyroll's applier if that's already sitting in your workflow. Chain several of these nodes together through the optional lora_stack input and each one appends itself to the pile.

The same Civitai-lookup trick as the rest of the pack rides along: pick a local LoRA from the dropdown, it BLAKE3-hashes the file, checks it against Civitai, and hands back trigger words and a preview if it recognizes it - same graceful "not found" fallback as every other loader here if it doesn't.

The inputs and outputs that matter

model_name is a dropdown of your local loras folder. lora_weight and clip_weight (both default 1.0, -10 to 10) are the split this node is named for - dial clip_weight down independently if a LoRA is overpowering your prompt adherence without touching how strongly it patches the model itself. bypass (default off) is a soft mute: flip it true and this entry drops out of the stack entirely without you rewiring anything - genuinely useful for A/B-testing which LoRA in a five-deep chain is actually doing the work. override_trigger_words and preview_images behave exactly like every other loader in this pack. lora_stack (optional) is where an upstream stacker plugs in.

Output: LORA_STACK carries forward to the next stacker or an applier node, civitai_trigger_words (LIST) and summary(Text) are the usual bonus pair.

How to install it

ComfyUI Manager - search "ComfyUI Easy Civitai (XTNodes)" and install - or clone manually and pip install -r requirements.txt, then restart.

Common issues & troubleshooting

Worth naming honestly: rgthree-comfy's Power Lora Loader does a version of this same job - stack LoRAs, surface trigger words - and it's the far more popular, actively maintained pack for it. If all you want is trigger-word convenience on a stack, that's a reasonable thing to have in mind. What XTNodes gives you instead is the paired Civitai-URL variant (CivitaiLoraLoaderStackedAdvanced) that can grab a LoRA straight from a model page URL, which rgthree doesn't do.

On reliability: this pack hasn't shipped a commit since September 2024, and its loader import is fragile in a specific way - if any of its five pip dependencies fails to install, that file's nodes vanish with zero error message, since the pack's own loader swallows the exception silently. If this node - or any loader from this pack - never shows up in your node search after installing, that's the first thing to check: re-run pip install -r requirements.txt and actually read what it printed.