FEEncLoraLoader

Most LoRA loaders are a widget, a dropdown, and a shrug. FEEncLoraLoader is the stock ComfyUI LoraLoader with two extras bolted on: it resolves short alias names through a map file the author calls the "EMP cache", and it gives you a third output with the LoRA's safetensors metadata. If you don't use either feature it just behaves like a normal loader, which makes it a safe drop-in.

The "Enc" in the name is the pack author's shorthand, and the practical difference is the exist_map.json file it reads from your loras folder. If models/loras/exist_map.json exists, it's used as a mapping between real filenames and short aliases: the dropdown shows you the friendly names, and behind the scenes the loader walks the map until it lands on a real file. No map file, no problem - the dropdown just shows your actual .safetensors filenames and everything works like a vanilla loader. It degrades gracefully, which is more than most of this pack's nodes can claim.

The inputs and outputs that matter

The wiring is exactly what you expect from a LoRA loader:

• model (MODEL) and clip (CLIP) - come from your checkpoint loader.
• lora_name - the dropdown, populated from your loras folder (or the alias map).
• strength_model and strength_clip - both default to 1.0, range −20 to 20 in 0.01 steps. For characters and styles you'll usually land between 0.5 and 0.8; 1.0 is often too hot. You can go negative if a LoRA is dragging the image somewhere you don't want.

Outputs are MODEL and CLIP - wire them into your sampler as usual - plus the party piece, EXTRA. That's a *-typed output carrying whatever training metadata is embedded in the safetensors header (ss_* fields, base model, sometimes trained words). Two useful things to do with it: stick it into FEExtraInfoAdd so the metadata rides along in your saved PNG, or just peek at it in a debug view to see what a downloaded LoRA was actually trained on before you fight its trigger words. A LoRA's own trained tags come out of the same header the CivitAI page shows, so EXTRA is your way to read them locally without a browser tab.

Installing

This is part of the fexli/fexli-util-node-comfyui pack. Easiest via ComfyUI Manager: Manager → Install Custom Node → search fexli-util-node-comfyui → Install. Or by hand:

cd ComfyUI/custom_nodes
git clone https://github.com/fexli/fexli-util-node-comfyui.git
cd fexli-util-node-comfyui
pip install -r requirements.txt

Then restart ComfyUI. The pack's requirements are just pycryptodome and sseclient-py - no model downloads, no torch build. The alias map is optional; if you want it, drop an exist_map.json into models/loras and restart.

Common issues

• LoRA loads but does nothing. Same failure mode as every loader: the file is on an incompatible base model. Illustrious and Pony both descend from SDXL, but a Flux LoRA won't touch an SDXL checkpoint, and ComfyUI won't warn you - it just ignores it. Check the ss_sd_model_name in EXTRA against what you loaded.
• Metadata shows up empty. EXTRA reads the safetensors header; if the file was converted or stripped it can be {}. That's a file problem, not a node problem.
• Failed to load lora in the console with the loader raising - usually a corrupted or truncated download. Redownload the file.

It's not the most famous loader in the ecosystem - rgthree's Power Lora Loader is the one people actually reach for when they want trigger words at a right-click. But if you want a stock-compatible loader that also hands you the metadata as data, this one is quietly useful.