LoRA Loader Pixaroma

What it does

Add as many LoRAs as you want, each on its own line with its own on/off switch and strength. Wire your model (and clip) in the top, and the modified model and clip come out. You can chain several of these nodes if you like.

Model and CLIP

A LoRA can change two things: the image model (the drawing side, the required model input) and CLIP (the part that reads your prompt words, the optional clip input). Most of a LoRA's look comes from the model side, so it works with only model connected. Connect clip too (checkpoint clip through this node and on to your text encode) when you want the LoRA to also tune how its trigger words are read - it matters most for trigger-word LoRAs. Leave clip unwired only in model-only setups.

Each row

• The name box opens a searchable list of your LoRAs (grouped by subfolder).
• The strength box: type a number, or use the small up/down arrows.
• The i button opens the info panel.
• The switch on the right turns that LoRA on or off; an off row dims.

Trigger words

Pick a LoRA and its trigger word is ticked for you straight away, so a LoRA that needs one just works. The ticked words come out of the triggers output as plain text you can wire into your prompt.

Only the real trigger is ticked: the words the maker declared in the file, or the ones saved from Civitai. Many LoRAs also carry a long list of tags from their training pictures, and those are offered in the panel but never ticked for you, because they are labels rather than the word the LoRA is waiting for.

Click the i on a row to see everything it has and change the picks. Un-tick a word and it stays un-ticked, even after a run and after closing the panel. If a LoRA has no words in its file you can type your own in the box at the bottom, or use the optional Civitai button to look them up online (only when you click it) and save them for next time. All of this is read from the file, so it works offline.

Your own trigger words are remembered

A word you type is kept for that LoRA, not just for the row you typed it on. Use the same LoRA in another row, another node or another workflow and your words are waiting in its panel. Swap a row to a different LoRA and back, and they come back too.

They are stored on this computer, alongside your other Pixaroma settings, so they survive an update and nothing is written into your models folder. Tap a word to send it to the triggers output, the same as any other word: being remembered means it is offered again, not that it is switched on for you. The small x on a word forgets it for that LoRA everywhere.

If you move a LoRA to another folder

The row still points at the old place, so its name shows in red. Click the name and pick the LoRA from its new folder: the words you ticked and typed on that row come along, and your typed words are remembered under the new place the next time you open its panel.

Words saved from Civitai live in a small .civitai.info file next to the LoRA, so move that file with it, or press the Civitai button in the panel once to look them up again. Your own picture stays with the old place, so set it again. Switching a row to a different LoRA still starts fresh, even when the two files share a name.

Use your own picture

The little picture at the top left of the info panel is the LoRA's preview. Click it to choose an image from your computer, or drop one onto it, or copy an image and press Ctrl+V over the panel. Handy for a LoRA you trained yourself, one that is not on Civitai, or simply when you would rather see your own example than the one that came with it.

Your picture wins over everything else, including a later Civitai lookup, and the small x on the corner removes it and brings the automatic one back. Nothing is overwritten: the picture is kept on this computer alongside your other Pixaroma settings, never written into your models folder, so it survives an update and works even when your models sit on a read-only or network drive. It follows the LoRA, so every row and every workflow using that LoRA shows it.

If you would rather see no pictures at all, for instance while recording, switch off Show preview thumbnails in the gear. The picture box then disappears from the panel entirely, and any pictures you have set are kept safe for when you switch it back on.

Comparing LoRAs side by side

XY Plot Pixaroma can drive this node. In its picker your rows show up under this node's name, counted from the top: LoRA 1 file tries a different LoRA file in every square, LoRA 1 strength sweeps that row's weight, LoRA 2 file is the next row down, and so on (there is a separate clip strength entry when you have split model and clip in the gear). The list you then tick is every LoRA on your computer, because the point is to try ones you have not loaded yet, and each tick makes one more square rather than stacking them together. An axis drives ONE row. The row being compared is switched on for every square, and the other rows stay exactly as you left them, which means a second switched-on LoRA is applied on top of every square and can hide what the compared row is doing. Switch off anything you are not comparing, and XY Plot will name the row in an orange line under its picker when one would get in the way. To compare two LoRAs in every combination instead, put one on the X axis and the other on Y. One thing to know when you compare LoRA FILES: that row's ticked trigger words are left out of every square, because they belonged to one particular LoRA and keeping them would give a single square wording none of the others have. Other rows keep their words as normal, and anything you want in every square belongs in your prompt.

Buttons and settings

Add LoRA, the all on/off switch, and the gear sit in the middle of the node. The gear opens the settings (default strength, step size, separate model and clip strengths, the trigger separator, the Civitai button, thumbnails, your Civitai account, and the highlight colour). Right-click a row to move it, duplicate it, or remove it.

When Civitai cannot find your LoRA

Civitai hides adult-rated models from anyone who has not signed in, and it hides them by answering exactly the same way it answers for a file it has never seen. So the lookup reports "not found" and there is no way to tell the two apart. If that is happening to you, the Civitai account section in the gear has the two things that fix it.

Add key takes an API key from your Civitai account (on their site: your profile menu, Account settings, then API Keys). The lookup then asks as you rather than as a stranger, so anything your own Civitai settings let you see comes back normally. Ask this site first switches between civitai.com and civitai.red, their unrestricted address - try Unrestricted if a LoRA still is not found. Whichever you choose, the other address is asked as well when the first one comes back with nothing, so nothing you could reach before stops working. Your key is sent to whichever of the two is being asked, since that is what lets either of them show you a hidden model.

Your key is kept on this computer, in ComfyUI's own user folder, and is never written into a workflow - so sharing a workflow, or an image with a workflow inside it, cannot give your key away. It is not shown again after you save it either: the settings only show the last four characters so you can tell which key is loaded.

Allow adult preview images is separate and off to begin with. A LoRA whose example pictures are all adult shows no picture at all unless you turn this on.

LoRA memory use

In the gear. Standard keeps only the last used LoRA file in memory between runs, the way ComfyUI itself does - the balanced default. Fast keeps the whole stack loaded for the quickest re-runs, which can hold a lot of memory with several large LoRAs. Lowest keeps nothing and re-reads the files each run - the best choice on a machine with limited memory. If a LoRA's file has been renamed or removed, its row shows a red warning and it is skipped until you pick the file again.