Does a LoRA slow down generation?

Barely: it adds a few small matrix multiplications per step. Loading a LoRA the first time takes a second or two; after that the cost is negligible.

Can I use LoRAs on this site?

Not yet: the public tools run fixed models. Members' txt2img pages may expose specific LoRAs in the future.

What does the number in mean?

The strength with which the LoRA's changes are applied: 1.0 is what the author trained, 0.5 is half the effect, values above 1.2 usually distort the image.

What is a LoRA in Stable Diffusion? Styles and characters

A LoRA (Low-Rank Adaptation) is a small file, usually 20 to 150 MB, that adds a style, a character, a product or a concept to an existing Stable Diffusion model without retraining it. You load it next to the base model, call it in the prompt with a weight, and combine several at once.

How a LoRA works

Fine-tuning a whole model means rewriting billions of weights and producing a 2 to 7 GB file. A LoRA freezes the model and trains only two small matrices per attention layer whose product approximates the change. The result captures "what is new" in a few megabytes and is applied on the fly, at generation time, with a multiplier. Ten to thirty images and a few minutes on a consumer GPU are enough to train one.

How to use a LoRA

Download a LoRA made for your base model: a SD 1.5 LoRA does not work with SDXL, and vice versa; SD 3.5 has its own.
Put the file in models/Lora of Automatic1111 (or the equivalent folder of ComfyUI, Forge, Fooocus).
Call it in the prompt: <lora:filename:0.8>. The number is the weight; 0.6 to 1.0 is the usual range.
Add the trigger word given by the author if there is one (a character name, a style token).
Adjust: too strong and every image looks the same, too weak and nothing changes.

Combining LoRAs

Several LoRAs can be stacked: a style at 0.7, a character at 0.8, a detail enhancer at 0.4. Keep the sum of weights around 1.5 to 2 at most, and fix the seed while you tune them, otherwise you cannot tell what each one does. LoRAs trained on the same concept fight each other; pick one.

LoRA, embeddings, checkpoints: which to choose?

Type	Size	Best for	Used how
Checkpoint (fine-tune)	2–7 GB	a whole aesthetic (realistic, anime)	replaces the base model
LoRA	20–150 MB	a style, character, object, pose	<lora:name:weight> in the prompt
Textual inversion (embedding)	10–100 KB	a single token, often a negative	the token name in the prompt

Most people run one good checkpoint and swap LoRAs per image.

Where to find LoRAs

Civitai and Hugging Face host tens of thousands, with sample images, trigger words and the base model they need. Check the licence: many forbid commercial use or the reproduction of real people. Our artist style comparison is a good alternative for styles: a name in the prompt often does what a style LoRA would, with nothing to download.