Ino Load Lora Model

LoRA loading in ComfyUI is normally a dropdown with a strength slider, and that's fine until you want the graph to decide which LoRA to apply. InoLoadLoraModel is the path-string version: give it a model, give it a LoRA path, and it patches the model - model only, no CLIP. That last bit is the deliberate design choice, and it's exactly right for a lot of modern setups.

Native Load LoRA applies a LoRA to both the model and the CLIP text encoder, which you need when the LoRA was trained with its own tokens baked into the text encoder. Model-only loading (what LoraLoaderModelOnly does natively) is what you want when you're using a LoRA that only touches the denoiser - increasingly common on newer architectures, and it's also the lighter-weight option when you don't need the text-encoder patch. If you do need both, the pack's sibling Ino Load Lora Clip Model exists for exactly that.

How it works

Required: model_path (the LoRA file, relative to ComfyUI/models/loras/), model (a MODEL from your base loader), and enabled. Optional: strength_model (default 1.0, the LoRA weight - the old "0.5 to 0.8 is safer" advice was SDXL-era, so on newer bases don't be shy about 1.0).

The two outputs that make this node special:

• name - the LoRA's filename without extension, derived from the path.
• trigger_word - the author's chosen trigger, guessed by taking everything before the first underscore in the filename. It's a heuristic, so it's only as good as the file's naming, but for auto-pipeline work it's a cheap way to build prompts dynamically: wire trigger_word into a string concat and then into your text encoder.

Plus the usual success, message, and the patched model output to continue the chain.

Installing it

Part of the 125+-node ComfyUI-InoNodes pack by nobandegani:

cd ComfyUI/custom_nodes
git clone https://github.com/nobandegani/comfyui_ino_nodes
cd comfyui_ino_nodes
pip install -r requirements.txt

Or search "ComfyUI Ino Nodes" in ComfyUI Manager, install, restart. V3 schema pack - keep ComfyUI current.

Common issues

Two things trip people up. First, model_path must be relative to models/loras/, matching what ComfyUI's own LoRA loader expects - an absolute path or a path into the wrong subfolder gets you an exception in message. Second, the trigger word is derived from the filename, not read from the file's metadata. my_character_v2.safetensors gives you my as the trigger, which is almost certainly wrong - rename the file to start with the real trigger word if you rely on this output. LoRAs are architecture-bound (an SDXL LoRA does nothing on Flux), so double-check the LoRA actually matches the model family you loaded. And if success is False with "not enabled," you left the toggle off - every Ino node carries one.