Extract prompt from LoRA tag

Here's a workflow problem nobody sells you a node for: you load a LoRA and then have to type its description into the prompt by hand. LF_ExtractPromptFromLoraTag automates that step - if your LoRA files are named the way this node expects, it splits the filename apart and hands you back the keywords, pre-formatted and ready to wire into your prompt.

The mechanism lives in a naming convention. If your LoRA is called portrait_SEP_softlight_SEP_bokeh.safetensors, the node uses that SEP separator to break the name into its meaningful chunks and returns portrait, softlight, bokeh as a usable keyword string. It's the same trick the author described when he launched the refactor of this pack - "extract keywords from a LoRA tag, if the files are named accordingly." The whole node is basically a bet on tidy naming, and it pays off if you're willing to rename your files.

How it works

Two inputs. tag is the LoRA tag to convert - multiline, so it can hold a batch. separator defaults to SEP; change it if your convention uses something else, like _ or +. There's no hidden parsing magic, just a split-and-strip.

The four outputs are designed for different wiring patterns:

• keywords - the combined keyword string, comma-separated. This is the one you probably feed into your prompt.
• keywords_count - how many keywords ended up in there.
• keywords_list - the keywords as a list, one per element, for nodes that iterate.
• keywords_count_list - the counts per tag as a list, when you passed several tags in.

If you're batch-testing LoRAs - loading five of them and wanting each one's trigger words in the prompt without hand-editing - this node plus a loop is the clean solution. It also pairs naturally with the pack's other "work from a filename" utilities, because this suite is big on driving the graph from well-named files.

Installing it

Part of LF Nodes, so:

cd ComfyUI/custom_nodes
git clone https://github.com/lucafoscili/comfyui-lf

Restart, or use ComfyUI Manager and search LF Nodes. No model downloads; the pack deps are minimal (numpy, Pillow, opencv-python, requests, torch).

Gotchas

The whole thing depends on your filenames being consistent. If your LoRAs are named myrandomname_v1, there are no keywords to extract and you'll get one big blob. The node is only as smart as your naming scheme - which is exactly what the tooltips say, and worth knowing before you build a workflow around it.

And the pack note: comfyui-lf is legacy (frozen Feb 2025); active development moved to lucafoscili/lf-nodes. The frozen version works, but the new repo is where this class of utility keeps evolving.