WeiLin Lora Stack

This is the pared-down member of the WeiLin-Comfyui-Tools family: just the LoRA stack, none of the tag editor. The pack's two prompt nodes (WeiLinPromptUI and WeiLinPromptUIWithoutLora) both come with a full tag-management UI wrapped around them, which is overkill if all you actually want is a place to stack a few LoRAs onto a model and CLIP before they hit your sampler. This node is that, and only that - the author added it specifically for people who wanted the LoRA manager without also getting a prompt editor they didn't ask for.

Why you'd reach for it. Stacking multiple LoRAs is a solved problem in ComfyUI several times over - core's LoraLoader chained a few times, rgthree's Power Lora Loader (the community's default answer, with per-entry toggles and CivitAI trigger-word lookups baked in), or this. What WeiLin's version brings to that crowded field is its LoRA manager panel: browse installed LoRAs with previews and metadata pulled from CivitAI, click to add, and it writes out using the pack's own tag syntax, <wlr:LoraName:modelWeight:textWeight> - one entry per LoRA, each with independent model and CLIP/text weights, the same two-weight split every serious LoRA loader gives you. If you're already inside WeiLin's ecosystem for tag management elsewhere in your workflow, this keeps the LoRA browsing experience consistent instead of switching to a different pack's UI mid-graph. If you're not, it's a perfectly reasonable standalone LoRA stack node on its own merits - worth trying alongside rgthree's version and core's chained loaders to see which UI actually fits how you work, since none of them are meaningfully different in what they do to your model weights.

How it works. You feed it a model and CLIP, open its LoRA manager, add however many LoRAs you want with their weights, and the node parses the resulting lora_str tag list and applies each one to the incoming model and CLIP in sequence - mechanically the same thing any LoRA stack does: freeze the base weights, patch in each adapter's low-rank matrices at the strength you set. temp_lora_str is the same kind of scratch field its siblings carry - editor bookkeeping, not something you need to write into by hand.

Inputs and outputs. clip and model are both required - unlike the two prompt nodes where CLIP and model are optional extras, here they're the whole point, since the node exists to patch and re-emit them. lora_str (optional) is where the parsed LoRA-tag list lives once you've added entries through the manager UI. Outputs are just CLIP and MODEL, LoRA-patched and ready to feed into a CLIP Text Encode and your sampler's model input respectively. Unlike its two siblings, this one is not flagged as an output node - it's a pure mid-graph utility, meant to sit between your checkpoint loader and everything downstream, not to be a leaf ComfyUI force-executes on its own.

Install. ComfyUI Manager, search "WeiLin-Comfyui-Tools" - installing the pack gets you all three nodes, there's no way to grab just this one. Manually: cd ComfyUI/custom_nodes && git clone https://github.com/weilin9999/WeiLin-Comfyui-Tools, then restart. Nothing extra to download; it works against whatever LoRAs you already have in your models folder.

Common issues. If a stacked LoRA isn't doing anything, run the standard checklist before assuming the node is broken: wrong base-model family for that LoRA (an SDXL LoRA does nothing on Flux, and even within the SDXL family Pony/Illustrious/vanilla LoRAs aren't reliably cross-compatible despite all being selectable), a missing trigger word in your prompt, or a weight that's too low for that particular LoRA - the old "0.5 to 0.8" rule of thumb is SDXL-era folklore and plenty of newer LoRAs genuinely want full strength. As with the rest of the pack, install or update it and do a proper restart rather than a Manager "reload" - the pack's server-side routes don't always pick up on a soft reload. If you're stacking several LoRAs and getting black images or NaN errors, that's a known general failure mode of LoRA stacking regardless of which loader you use - try reordering with the larger LoRAs first, or isolate which one is the culprit by disabling entries one at a time.