凤希AI - LoRA名称选择
One LoRA name, five loaders, no typos
- lora_name
The tiny node that saves you from string bugs
A LoRA loader's lora_name field is a dropdown you pick from. That's fine until you need the same LoRA in three places - a model-only loader, a strength-scheduled branch, a second sampler - and now you're maintaining three copies of a filename that has to match byte for byte, including whatever Chinese characters or version suffix the author put in it.
FxAiLoraNames (凤希AI - LoRA名称选择) turns that filename into a wire. Pick it once, feed every loader from one node, change it in one place.
How it works
It's about as thin as a node gets: the input is built from folder_paths.get_filename_list("loras") - ComfyUI's own index of your models/loras directory, subfolders included, so the dropdown looks exactly like the one on a stock LoRA Loader - and the function returns the chosen string unchanged.
There's one input, lora_name, and one output, lora_name, typed as a wildcard so it can be plugged into any loader's name input without a type fight.
Don't read anything mystical into it. It's a value node: the whole contribution is separating which LoRA from where the LoRA is applied, which is the same reason people use a core Primitive for seeds.
Wiring it
On each LoRA Loader, right-click the lora_name widget and convert it to an input (or drag off the widget on newer frontends), then connect this node's output. One dropdown now drives all of them, and the graph shows you at a glance that they're all the same file rather than three strings that happen to look alike.
The case that actually bites people is LoRA strength sweeps and two-pass workflows: same LoRA at a different weight in each pass. Two loaders, one name node, and you vary only the strengths - no risk of discovering at the end of a long run that the second pass silently loaded a different version of the file.
Installing
cd ComfyUI/custom_nodes
git clone https://github.com/fxai666/fxai-toolkit
Restart ComfyUI afterwards - and this node is the one place in this article where the restart genuinely matters, because the LoRA list is read when the node definition is built. A LoRA you drop into models/loras while ComfyUI is running won't appear in the dropdown until you refresh node definitions (R on some frontends, or a restart). Same for deleting one.
ComfyUI Manager works too: search fxai-toolkit (publisher fxai). The pack pip-installs soundfile and psutil on import. Nothing to download for this node.
Where it bites
It validates against your machine, not the author's. Because lora_name is an enum built from your local file list, a workflow shared from someone else fails validation when it references a LoRA you don't own - value not in list, before anything runs. That's arguably correct behaviour (a silent fallback to no-LoRA would be worse), but it means shared workflows with LoRA bells on them always need a file-collecting step first. Put the LoRA in the same subfolder path it expects and the name matches again.
Case and subfolder are part of the name. Style/foo.safetensors and style/foo.safetensors are different strings, and Windows will forgive a case difference that later bites you on a Linux box. Copy the exact string the dropdown gives you.
It picks a name; it doesn't load anything. No trigger words, no strength, no applying. If you also want the trigger words to follow the LoRA around, that's a different problem - this node has no opinion.
Doesn't appear in the menu. This pack imports each node separately and logs ❌[凤希AI] 节点加载失败 on the console for anything that fails, so one broken module quietly removes one node. If the 凤希AI/资源 category is missing entirely, suspect a missing pip dependency rather than the node itself.
Inputs (1)
| Name | Type | Default | Description |
|---|---|---|---|
| lora_name | COMBO | 0 options: |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| lora_name | * | — |