Load LoRA (Model Manager)

A LoRA is a small adapter file that patches a checkpoint's behavior without replacing it - the way you add a specific character or style without a new 7 GB download. Load LoRA (Model Manager) is this pack's version of the core Load LoRA node: it pulls the LoRA from your Model Manager server, applies it to the model, and hands you back the patched model plus a little record of what it just did.

What you're setting

The important inputs are exactly the ones you'd expect from any LoRA loader:

• model - the MODEL coming out of your checkpoint or diffusion-model loader.
• lora_name - the dropdown of LoRAs from the server's LoRA category (modelId@versionId:name entries; shows (not connected) until you authenticate).
• strength_model and strength_clip - the two weights that decide how hard the LoRA bites. Defaults are 1.0, and 0.5–0.8 is the usual place people actually land; you can go negative (to subtract a style) or above 1.0 if you're feeling brave.
• Optional clip input, so it can patch the text encoder alongside the model.
• Optional lora_info input - a place to chain in info from another loader if you're building a stack.

Outputs: model and clip (the patched pair, wire straight into your sampler and CLIP Text Encode), plus lora_info (the record of what was applied, for uploads and inspection) and model_ref (the raw lora_name string, handy for routing).

How it works

First run, it streams the LoRA down with a progress bar and caches it in ComfyUI/models/model_manager_cache/loras/. Then it loads the weights and applies them through ComfyUI's standard LoRA machinery. Two nice touches under the hood:

• The node keeps the last loaded LoRA in memory and only re-reads the file when the selection changes - repeated runs don't re-hit disk.
• The IS_CHANGED hook tracks the connection version, so the dropdown stays fresh when the server's library changes.

Install

ComfyUI Manager (search comfyui-model-manager-nodes) or:

cd ComfyUI/custom_nodes
git clone https://github.com/blackpaw-studio/comfyui-model-manager-nodes
cd comfyui-model-manager-nodes
pip install -r requirements.txt

Restart, connect via the Connect to Model Manager button (URL + API key, saved to config.yaml; MODEL_MANAGER_API_URL / MODEL_MANAGER_API_KEY env vars override). Light dependencies: requests and pyyaml.

Troubleshooting

• (not connected) in the dropdown - connect first; a 401 means the key's wrong or expired.
• LoRA silently not affecting output. Check strength_model - at 0 or near it, the LoRA does nothing. Also confirm the LoRA's base model matches your checkpoint (an Illustrious LoRA on a Flux checkpoint is a mismatch, not a failure).
• Downloading every run? No - it caches. If you cleared the cache with Clear Cache (Model Manager), expect one re-download per LoRA.
• Need the file in ComfyUI's normal loras folder instead? That's what Download LoRA (Model Manager) is for - it saves straight into ComfyUI/models/loras so any stock loader can see it.

Single-LoRA needs are covered here. If you're stacking three LoRAs per image, the multi-loader sibling is the one you actually want.