💊 Easy LoRA Character Loader

Character LoRAs are where the "where was I?" prompt problem gets worst. The LoRA file loads fine, but its trigger word lives in a Civitai tab somewhere, the ideal strength is a number you tuned once and forgot, and the clothing or expression variations are tags you have to re-derive every session. Easy LoRA Character Loader, from Comfy Cabinet, pins all of that to the LoRA itself.

It's the flagship of the pack's five LoRA nodes, and the most opinionated about how prompts should be organized. It's built for character LoRAs specifically (the Civitai main tag Character), and it assumes you think of a character as having four separate prompt dimensions: the character itself, clothing, no_clothing, and expression. Each of those gets its own saved block of tags and its own output socket.

How it works

Mechanically it's a standard LoRA loader with a memory layer. It takes your model and clip, applies the LoRA via ComfyUI's own LoraLoader, and returns the patched MODEL and CLIP just like the core node would. The extra work happens in the prompt block outputs: each one (character, clothing, no_clothing, expression) checks whether you typed something on the canvas, and if you didn't, falls back to the saved config for that LoRA stored in a local SQLite database (lora_configs.db in the pack folder). Typed value wins; saved value is the fallback. That's the pattern across all five specialized loaders.

Two conveniences are worth calling out. The lora_name dropdown is category-filtered on the client side, so the Character node lists character LoRAs rather than dumping your entire loras folder at you (if nothing matches your filter it falls back to the full list rather than stranding you). And when you pick a LoRA from the dropdown, it auto-populates the strength widgets and prompt blocks from the saved config - which is the moment the whole design clicks: select the character, and the node remembers how you wanted it.

The strengths work exactly like the core LoraLoader: strength_model and strength_clip, default 1.0, range -20 to 20, and setting both to 0 passes the model and clip through unchanged, which is a clean way to bypass a LoRA without rewiring the graph.

What you set

The required inputs are model, clip, lora_name, and the two strength floats - wire model and clip in from your checkpoint loader, pick the LoRA. The four optional multiline fields (character, clothing, no_clothing, expression) are where you put trigger words and tags. On output you get MODEL and CLIP for the sampler chain plus the four STRING blocks, which you can feed into your text encoder - either all four stitched into one prompt or routed independently to a regional conditioner.

The one field that confuses people is no_clothing. It's the "without the outfit" trigger - some character LoRAs train a tag for the character sans outfit, and that's the slot for it. If your LoRA has no such concept, leave it blank and it just won't contribute.

Getting it installed and configured

Same one-command install as the rest of the pack:

cd ComfyUI/custom_nodes
git clone https://github.com/Zed93/Comfy-Cabinet.git

Restart ComfyUI, or install via ComfyUI Manager by searching "Comfy Cabinet". The only dependency is aiohttp, which ComfyUI already ships. The config UI lives behind the 🗄️ Easy Config button in ComfyUI's top menu - open the LoRA config tool, browse the gallery (previews, tags, trigger words sync from Civitai if you paste in a key; the pack supports both civitai.com and civitai.red), tune strengths and blocks, and hit Send to ComfyUI.

Where people get burned

Your configs are keyed by the exact LoRA filename, so renaming a LoRA orphans its saved triggers and strengths - keep filenames stable. The Civitai metadata sync is bring-your-own-key: without a key you lose the pretty gallery and auto-trigger words, but generation is unaffected. And the pack is release-candidate software tested against ComfyUI 0.30.1, so if a node or the Easy Config button misbehaves after updating, restart first. One thing to know: this node is for character LoRAs only. If you're grabbing a style LoRA or a clothing LoRA, there's a matching specialized loader (or the catch-all Basic one) that fits better.