# Using LoRAs (/docs/models-loras/using-loras)













## What's a LoRA? [#whats-a-lora]

A LoRA is a small style or character add-on that you stack on top of an
image model. They add a particular look, character, or effect without
having to retrain the whole model. They're tiny compared to a full
checkpoint, usually 50 to 300 MB.

<Video youtubeId="kdfXTCBgriU" title="Using LoRAs" />

## 1. Open the LoRA library [#1-open-the-lora-library]

Click **Local Engine*&#x2A; on the bottom system strip of the left rail to open the engine popover, then click &#x2A;*"Model Manager"**. Inside the Model Manager, switch to the **LoRAs** tab.

<img alt="The Local Engine item on the left rail, opening the engine popover with Model Manager" src="__img0" />

<img alt="Model Manager with the LoRAs tab selected" src="__img1" />

## 2. Add a LoRA [#2-add-a-lora]

Click **+ Add LoRA**. You'll see two tabs:

* **From URL**: paste a direct link to a `.safetensors` file from
  HuggingFace or CivitAI. SimpliGen downloads it for you.
* **Upload File**: pick a `.safetensors` file you already have on your
  computer.

<img alt="Add LoRA modal, From URL tab" src="__img2" />

<img alt="Add LoRA modal, Upload File tab" src="__img3" />

## 3. Fill in the details [#3-fill-in-the-details]

You'll be asked for a few things:

* **Display Name**: what you'll see in the picker later. Pick something
  you'll recognize, like *Cinematic Realism v2*.
* **Base Model**: the model architecture this LoRA was trained for
  (SDXL, FLUX, Wan 2.2, etc.). &#x2A;*This matters.** A LoRA trained for SDXL
  won't work on FLUX, and SimpliGen will warn you if you try to use a
  mismatched LoRA.
* **Trigger Words*&#x2A; &#x2A;(optional)*: comma-separated words that activate
  the LoRA's effect. They get added to your prompt automatically. Check
  the LoRA's page on HuggingFace/CivitAI for these.
* **Default Strength**: how strongly the LoRA influences the result.
  `1.0` is the standard starting point; `0.7` is gentler.

Hit **Save** (or **Download & Add** if you used the URL tab) and
SimpliGen adds it to your library.

## 4. Use it in a generation [#4-use-it-in-a-generation]

Open the prompt builder and pick a preset that matches the LoRA's base
model. Scroll down to the &#x2A;*LoRAs (optional)*&#x2A; section. Click **+ Add LoRA**, pick
your LoRA from the list, and tweak the strength if you want. You can
stack several LoRAs together.

<img alt="LoRA stack selector in the prompt builder with a LoRA active" src="__img4" />

Generate as normal. The LoRA's trigger words (if you set any) get
mixed into the prompt automatically.

## Tips that save pain [#tips-that-save-pain]

<Callout type="tip" title="Match base models">
  A LoRA's **Base Model*&#x2A; has to match the preset's model. SimpliGen
  shows a yellow &#x2A;"Built for X"* warning when the LoRA and preset
  don't line up. You can override it ("Use anyway"), but the
  result will usually be poor.
</Callout>

<Callout type="info" title="Where to put trigger words">
  Trigger words go in your **prompt**, not in the Display Name field.
  If you set them when adding the LoRA, SimpliGen handles this for
  you automatically.
</Callout>

<Callout type="info" title="Strength sweet spot">
  Most LoRAs land between **0.7 and 1.0**. Push higher for a stronger
  effect, lower if the LoRA is overpowering your prompt.
</Callout>

## Troubleshooting [#troubleshooting]

**"My LoRA doesn't show up in the picker."**
Check the Base Model filter at the top of the LoRA tab; it might be
filtering yours out. Also try searching by name in the search box.

**"The result looks the same as without the LoRA."**
Two usual causes: strength is too low (try `1.0`), or the trigger words
aren't in your prompt. Open the LoRA's page on HuggingFace or CivitAI
and check what trigger words it expects.

**"Generation fails right after I add a LoRA."**
The file is probably corrupted, or it's for a different model
architecture than what you set. Try re-downloading, or try a different
LoRA to confirm.

***

Still stuck? Hop into [our Discord](https://discord.gg/v3MkTUAUN6). We
help out there daily.