LoRAs — small fine-tuning models for Stable Diffusion — remain one of the strongest reasons to run Stable Diffusion locally rather than use a hosted generator. A LoRA is a few hundred megabytes at most, loads on top of a base checkpoint, and teaches it a specific character, style or concept the base model never saw.
The extensive variety of LoRAs created and shared by the community allows users to easily generate images of specific characters, styles, or other concepts not included in the original training data.
What is Fooocus?
Fooocus is a new interface for Stable Diffusion that offers a simplified, user-friendly experience for creating stunning images. It requires minimal technical knowledge and avoids complex settings. We detail the installation and usage of Fooocus in our introductory article.
Finding and Adding LoRAs
Where to Find LoRAs
The best resource for finding LoRAs is Civitai, a community sharing site where creators post their models, LoRAs, and prompts. Visit Civitai, click on "Models" at the top left, then use the filter to select "LoRA" to limit your search. Since Fooocus works exclusively with SDXL and its derived models, you can also check the "SDXL" box to find compatible LoRAs. Once you find a LoRA of interest, download it using the "Download" button on the right side of the LoRA page.
Placing the LoRA File
To add a LoRA to Fooocus, place the downloaded file in the /models/loras directory of your Fooocus installation. On Windows, Mac, or Linux, this directory is located on your hard drive where Fooocus is installed. Move the LoRA file (usually a .safetensor file) into this folder.
Adding LoRAs on Google Colab
If you are using the hosted version on Google Colab, follow the specific procedure for adding LoRAs in that environment.
Using a LoRA in Fooocus
Activating the LoRA
To activate one or more LoRAs in Fooocus, open the "Advanced" interface by checking the box of the same name at the bottom of the screen, below the prompt field. In the menu that appears on the right, select the "Model" tab. Under the "LoRAs" section, you can choose up to five LoRAs to use for image generation. If you don't see the newly added LoRA, click the "🔁 Refresh All Files" button at the bottom of the column to reload the list of available files.
All selected LoRAs will be used in the image generation process. Unlike using LoRAs in Automatic1111, there's no need to use special syntax in your prompt to utilize a LoRA. However, if the LoRA has specific activation keywords, you must include them in the prompt to properly activate the LoRA.
Adjusting LoRA Strength
For each of the five selectable LoRAs, you can set a weight to determine its influence on the image. The recommended weight generally varies between 0.5 and 1, but each LoRA is different and may have a different recommended strength. A weight too low will make the LoRA's influence imperceptible, while a weight too high will reduce the model's flexibility and degrade image quality. Check the LoRA's details or example prompts to get an idea of the appropriate weight, but you may need to experiment to find the best strength for your needs.
Examples of Using LoRAs in Fooocus
Tim Burton (Art Style)
- Prompt: "A pixie fairy in a Tim Burton-inspired wonderland. Around her, the landscape transforms into a surreal dreamscape, trees with twisted branches, each bearing glowing fruits. The pixies's eyes reflect the glow, holding a mixture of determination and curiosity, Tim Burton Style"
Rorschach - Watchmen
- Prompt: "Cinematic photo, Rorschach from the Watchmen, man with a hat, vintage ice cream store, holding an ice cream, 35mm photograph, film, bokeh, professional, 4k, highly detailed, Rorschach1024"
SDXL Enhancer
- Prompt: "Full-bodied portrait, cute and adorable cartoon white rabbit baby wearing a gold jaguar print hoodie and silver sunglasses, fantasy, dreamlike, surrealism, super cute"
As you can see, using LoRAs with Fooocus is both simple and powerful. These mini-models allow you to easily alter generated images to incorporate specific styles, characters, or interesting visual effects. Each LoRA is unique, and you will need to experiment to understand its potential fully. The weight used significantly impacts the result and can either enhance or ruin the image quality.
I hope this article has inspired you to try LoRAs with Fooocus to create new images. If so, feel free to join us on X and share your creations!
Stacking Multiple LoRAs
Fooocus gives you five LoRA slots, and all selected LoRAs apply together. That is powerful
and easy to overdo — combined weights are what usually wrecks an image rather than any
single LoRA being wrong.
A workable approach: keep the total weight across all active LoRAs near 1.0–1.5. One style
LoRA at 0.8 plus one character LoRA at 0.6 behaves predictably. Four LoRAs at 1.0 each will
generally produce a muddy, over-baked result no matter how good each one is on its own.
Add them one at a time. Get a single LoRA working at the weight you want, then introduce the
next — otherwise there is no way to attribute a bad result to a specific cause.
Why Your LoRA Seems to Do Nothing
It is not SDXL-based. Fooocus runs SDXL and its derivatives. A LoRA trained for SD 1.5
will either be ignored or produce noise. Filter for SDXL on the download page before you
download.
It is missing its trigger word. Many LoRAs only activate when a specific keyword appears
in the prompt. The keyword is listed on the LoRA's page, often as "trigger words" or inside
the example prompts. Without it, the LoRA loads and contributes almost nothing.
It is not in the list. New files are picked up when Fooocus rescans. Use the
"🔁 Refresh All Files" button at the bottom of the column rather than restarting.
The weight is too low. Below roughly 0.4 most LoRAs are imperceptible. Below 0.2 there is
effectively no effect at all.
The weight is too high. Above roughly 1.2 a LoRA starts overriding composition and
anatomy, producing burnt colours, duplicated limbs and rigid poses. If quality collapses as
you raise the weight, that is the ceiling for that LoRA.
It is in the wrong folder. The file belongs in models/loras, not models/checkpoints.
A LoRA placed with the checkpoints will appear in the wrong dropdown and fail to load.
Picking a Sensible Weight
| Weight | Effect |
|---|---|
| 0.2–0.4 | Barely present; useful for a hint of style |
| 0.5–0.8 | The reliable range for most style and concept LoRAs |
| 0.9–1.1 | Strong; typical for character LoRAs that must stay recognisable |
| 1.2+ | Usually degrades anatomy and colour; reserve for LoRAs that specifically ask for it |
Always check the LoRA's own page first — creators normally state a recommended weight, and it
beats guessing.
FAQ
Do I need special prompt syntax like <lora:name:0.8>?
No. That syntax belongs to Automatic1111. In Fooocus you select the LoRA and set its weight
in the Model tab, and the prompt only needs the LoRA's trigger word, if it has one.
How many LoRAs can I use at once?
Five. Keep the combined weight moderate rather than filling every slot at full strength.
Will SD 1.5 LoRAs work in Fooocus?
No. Fooocus is built on SDXL, so LoRAs must be SDXL-compatible.
Where exactly does the file go?
Into models/loras inside your Fooocus installation, as a .safetensors file.
Can I use LoRAs in the Google Colab version?
Yes, but the file has to be fetched into the Colab environment's models/loras directory
each session, since that storage does not persist between runs.
Do LoRAs slow generation down?
Marginally. Loading several at once adds a little overhead, but generation time is dominated
by resolution and step count.
Last reviewed and updated: July 2026.
Top comments (4)
er, are there supposed to be comparison images?
you know, the 2-6 ones?
Or am I missing something?
lol yeah ur not missing anything - there's supposed to be comparison pics but they're not here. Forgot to add them, my bad will fix sooooonn 💀
phew. I thought i was going completely insane...being half there already
Nooo LOL, you are not insane LMAO, im the one who forgets things XD