KurutAI LoRA Loader

The KurutAI LoRA Loader is this pack's version of ComfyUI's stock Load LoRA node, tuned for LoRAs from the KurutAI catalog. You feed it a model and a CLIP, tell it which KurutAI LoRA you want, and it hands back a modified model and CLIP plus the trigger words you're supposed to put in your prompt. The difference from the native node is the part you can't see: it verifies the file is actually on disk, and if it isn't, it knows exactly which one is missing and can pull it down for you.

How it works

Under the hood it looks the LoRA up in the pack's local download index using the lora_id, artifact_id, and sha256 you pass in. Found and hash-matching, it loads the weights through ComfyUI's own LoRA loader with your two strengths. Not found, it stops with a specific error: "This KurutAI LoRA is not installed. Queue again and confirm the download, or use Download in the KurutAI sidebar."

That error is the download mechanism, not a bug. Opening a workflow never downloads anything. The first time you queue a workflow that needs a missing KurutAI LoRA, ComfyUI asks for your consent, then fetches it - SHA-256 verified, with live progress and detailed console logging.

The inputs and outputs that matter

The metadata fields - lora_name, lora_id, artifact_id, sha256, trigger_words - usually arrive pre-filled because the KurutAI sidebar or Prompt Builder wrote them. You rarely touch them by hand. The two you'll actually adjust:

• strength_model - default 1.0, range −20 to 20. This does most of the work for SDXL-lineage character LoRAs (Illustrious, Pony, NoobAI).
• strength_clip - default 1.0, same range. You can usually leave it at 1 and forget about it.

Outputs: model and clip feed the KSampler and CLIP Text Encode exactly like the native node's. trigger_words is a STRING that passes straight through, handy for wiring into a prompt or text-concat node - or into the pack's own KurutAI Prompt node.

Install

ComfyUI Manager, search KurutAI for ComfyUI, install, restart. Or the old-fashioned way:

cd ComfyUI/custom_nodes
git clone https://github.com/KurutAI/KurutAI-ComfyUI.git

Restart, done. There are no extra Python packages - the pack's requirements.txt is literally a comment saying ComfyUI already provides aiohttp.

Where people get burned

Adding the node fresh and queueing with empty fields gives you that "not installed" error, because there's no lora_id to look up - the fields get filled when you pick something from the KurutAI sidebar. Also, if a file exists with the same name but a different hash, the pack won't silently overwrite it: it asks first and uses a numbered filename. Downloads land directly in your configured LoRA folder (Settings → KurutAI → Downloads → LoRA download folder), with no separate KurutAI model folder, and existing files are detected recursively across every configured folder.

If you already have a native Load LoRA wired the way you like and just want KurutAI to tell you the filename, that's the KurutAILoRASelector's job. This node is for when you want the whole verified-download pipeline in one box.