ComfyUI Extension: Finding LoRA

<h1 align="center">Finding LoRA — for ComfyUI</h1> <p align="center"> A LoRA loader with <strong>bookmarks</strong>, <strong>trigger-word storage and click-to-copy</strong>, <strong>fuzzy search</strong>, and a <strong>one-click button to drop another loader inline beside this one and wire it into the model chain automatically</strong>.<br> Stop scrolling a thousand-LoRA dropdown. Stop manually re-wiring whenever you want to stack another LoRA. </p> <p align="center"> <a href="https://buymeacoffee.com/lorasandlenses"><img src="https://img.shields.io/badge/Buy%20me%20a%20coffee-FFDD00?style=for-the-badge&logo=buy-me-a-coffee&logoColor=black" alt="Buy Me A Coffee"></a> </p> <p align="center"> <img src="node.png" alt="LoRA Loader (Finding LoRA) node — bookmarks bar, LoRA picker, bookmark/edit-trigger buttons, LoRA Strength slider, and Chain another LoRA Loader" width="520"> </p>

---

Why I built this

I have over a thousand LoRAs. ComfyUI's stock LoRA Loader makes me scroll a giant dropdown to find any of them, then if I want to stack another I have to drag out a second loader, wire its MODEL in, wire its MODEL out, and remember the trigger words. Every part of that workflow is friction I've felt hundreds of times.

This pack is what I wished the stock loader was: bookmark the LoRAs you actually use; type a few characters to find any LoRA in the folder; store the trigger words once and copy them with a click; chain another loader with a single button that wires itself into the model line for you. Each feature is small in isolation; together they make the most-used node in my workflow feel sharp instead of clumsy.

I haven't seen another LoRA-loader pack do this combination — bookmarks + trigger storage + click-to-copy + fuzzy search + auto-chaining — so here it is.

---

What's new vs the stock LoRA Loader

• No more horrible left/right chevron dropdowns. Both the LoRA picker and the bookmark picker open a proper modal on click. Default view is alphabetical with the current selection highlighted and scrolled into view, so a thousand-LoRA list lands you near where you were. Start typing to fuzzy-search — same picker, no separate search button.

• 📖 Bookmark button. One click bookmarks the currently-selected LoRA. The button label flips to 📕 Remove bookmark when the active LoRA is bookmarked.

• 📚 Bookmarks bar. A second clickable bar above the LoRA picker shows the active bookmark and opens a picker modal listing every bookmarked LoRA. Picking one sets the LoRA picker — the bookmark and the active LoRA always agree.

• ✏️ Trigger word storage. When you bookmark, you're optionally prompted for a trigger word or phrase. The trigger displays on a read-only line when a bookmarked LoRA is selected, and gets emitted as a STRING output you can wire into your prompt-encoding chain.

• 📋 Click-to-copy trigger. Click the displayed trigger row to copy the phrase straight to your clipboard — paste it directly into a CLIPTextEncode. Shows a quick toast confirming the copy.

• 🔗 Chain another LoRA Loader. A button at the bottom spawns a fresh copy of the node beside the current one and splices it into the model chain. Any downstream MODEL connections get re-routed through the new node — (upstream) → this node → new node → (former downstream). Stack as many LoRAs as you want without manual rewiring.

  <img src="chained-nodes.png" alt="Two Finding LoRA loaders side-by-side — the second was spawned by clicking 'Chain another LoRA Loader' on the first; the model line is wired through automatically" width="780">

• 🔍 Real fuzzy search built into the picker. Type in the picker modal to fzf-style fuzzy match across every LoRA in your loras/ folder. Substring matches always rank above scattered ones — typing snoys puts every name containing snoys at the top, before scattered subsequence matches sneak past on bonus points.

  <img src="lora-picker.png" alt="LoRA picker modal — alphabetical with current selection highlighted, type to fuzzy-search across the whole loras folder" width="640">

• 📚 Bookmark picker. Same modal UX, scoped to your saved favourites — type a few characters and it filters down instantly.

  <img src="bookmarks-picker.png" alt="Bookmarks picker modal — type to filter, ↑/↓ + Enter to apply" width="640">

• Bookmarks persist globally + sync live across nodes. Stored at <ComfyUI>/user/finding-lora/bookmarks.json. Add a bookmark on one node and every other Finding LoRA node on the canvas updates immediately — no restart, no refresh.

• All-canvas custom rendering. Every widget on the node — pickers, buttons, the strength slider — is canvas-drawn so text alignment is consistent and the node escapes the Vue-button-quirks that other custom-node packs run into. The strength widget keeps the standard interactions: click ◀ ▶ to step, drag the body to scrub, double-click for direct entry.

---

Quick start

1. Drop the comfyui-lora-FindingLora folder into ComfyUI/custom_nodes/.
2. Restart ComfyUI.
3. Add the LoRA Loader (Finding LoRA) node from the loaders category.
4. Wire it just like the stock LoRA Loader (Model Only) — MODEL in, MODEL out, plus a second trigger STRING output.

Then:

• Click the 🎛 LoRA: bar → modal opens at your current selection alphabetically. Click any name, or type kase and hit Enter, you've got character_kasey_v3.safetensors selected.
• Click 📖 Bookmark this LoRA → optional trigger word entry.
• Next time, click 📚 Bookmarks: to pick straight from your favourites.
• Click the trigger row to copy the trigger phrase to your clipboard.
• Click 🔗 Chain another LoRA Loader to drop another copy of the node beside this one and weave it into the model chain automatically.

---

Inputs and outputs

Required

• model (MODEL)
• lora_name (selected via the picker bar, but stored as a string)
• strength_model (FLOAT, default 1.0) — labelled LoRA Strength on the node face

Outputs

• model (MODEL) — with the LoRA applied
• trigger (STRING) — trigger word/phrase saved against the bookmarked LoRA, or empty string if not bookmarked. Wire into your prompt encoder via a string-concat node to auto-prepend trigger words to your prompt.

This is a model-only loader — no clip input or output, matching ComfyUI's stock LoraLoaderModelOnly for Klein 9B / Flux 2 / Wan / Z-image and any pipeline where CLIP isn't part of the LoRA chain.

---

UI reference

| Element | What it does | |---|---| | 📚 Bookmarks: | Click → picker modal listing all bookmarks. Type to filter. | | 🎛 LoRA: | Click → picker modal listing every LoRA. Default view is alphabetical scrolled to current. Type to fuzzy-search. ↑/↓ navigate, Enter selects, Esc cancels. | | 📖 Bookmark this LoRA | Bookmark the current LoRA. Prompts for an optional trigger word/phrase. | | 📕 Remove bookmark | Active LoRA is already bookmarked. Click to remove (with confirmation). | | ✏️ Edit Trigger Word | Edit the trigger word/phrase for the active bookmark. (When the active LoRA isn't bookmarked yet, the same button reads "Add trigger word (bookmark first)" and offers to bookmark on the spot.) | | 🔑 trigger row | Visible only when the active LoRA is bookmarked AND has a trigger saved. Click it to copy the trigger to your clipboard — a small 📋 hint icon sits at the right of the row. | | LoRA Strength | Click ◀ / ▶ to step ±0.01, drag the middle to scrub, double-click for direct entry. | | 🔗 Chain another LoRA Loader | Spawn another copy of the node beside this one and splice it into the model chain. Existing downstream MODEL connections are re-routed through the new node automatically. |

---

Search / fuzzy details

The search algorithm is a simplified fzf with a hard tier for substring matches:

• Tier 1 — substring match. Any filename literally containing the query as a contiguous substring scores in a band that no scattered match can reach. So if you type snoys, every filename containing snoys ranks above every filename where s/n/o/y/s is just scattered. Earlier-in-the-name and word-boundary substring matches get extra bonuses; gentle length penalty as the tie-breaker.
• Tier 2 — scattered subsequence (fzf-style). All characters in the query must appear in the target in order. Bonuses for consecutive chars, word-boundary starts (after _ - . / \ or whitespace) and camelCase boundaries. Length penalty as tie-breaker.

Top 200 results are shown to keep the UI fast even on huge libraries.

---

Storage

Bookmarks are stored as a single JSON file at <ComfyUI>/user/finding-lora/bookmarks.json. Format:

{
  "version": 1,
  "bookmarks": [
    { "lora_name": "character_kasey_v3.safetensors", "trigger": "kasey, blue eyes, freckles" },
    { "lora_name": "style_arcane_concept.safetensors", "trigger": "arcane style" }
  ]
}

Live editable on disk if you want to bulk-import or back up. Restart ComfyUI to pick up changes you make to the file directly.

---

Routes (for the curious)

The pack registers three HTTP routes on ComfyUI's server, namespaced under /finding-lora/ so they don't clash with anything else:

• GET /finding-lora/list — returns all bookmarks
• POST /finding-lora/add — { lora_name, trigger } adds or updates a bookmark
• POST /finding-lora/remove — { lora_name } removes a bookmark

Frontend talks to those, file storage stays server-side.

---

Support

If this saves you scroll-time, consider supporting development:

<a href="https://buymeacoffee.com/lorasandlenses"><img src="https://img.shields.io/badge/Buy%20me%20a%20coffee-FFDD00?style=for-the-badge&logo=buy-me-a-coffee&logoColor=black" alt="Buy Me A Coffee"></a>