Welcome to this tutorial on how to create wonderful images using Stable Diffusion with the help of LoRA models. In this post, we will have a closer look at how we can find, install, and use these so-called LoRA models. These models are instrumental in generating images with specific styles and themes. For instance, in this post, we'll explore how to create images featuring girls wearing traditional Chinese dresses. Let's dive in!
In our previous tutorials, we went to the website called Civit AI to find checkpoint models that are required to generate the desired images. These models come in various styles and sizes, with each offering a unique output. While some models are general purpose, others are more specialized. Let's explore how to find the LoRA models that suit our needs.
Here you can search for any LoRA model you like, I will be using the traditional chinese clothing model version 3 which can be found on Civit.ai website.
Once we've identified the desired LoRA model, we need to download and install it to our Stable Diffusion setup. Download the LoRA model that you want by simply clicking the download button on the page.
Once your download is complete you want to move the downloaded file in Lora folder, this folder can be found here; stable-diffusion-webui\models\Lora
These models are relatively compact, ranging from 50 to 200 megabytes, making them disk space-efficient.
With the LoRA model successfully installed, it's time to unleash its creative potential. We'll start by generating an image without the LoRA model activated to see the default output. Then, we'll explore how to activate the LoRA model, adjust its strength, and generate images with the desired style. You'll be amazed at the transformation!
Image without LoRA
Image with LoRA To clearly show the difference I used the img2img function in Stable Diffusion.
Below is a comprehensive guide available for you to explore about img2img in stable diffusion.
It's not as hard as you think, just follow these easy steps.
Click on the "Show/hide extra networks button", this can be found under the Generate button.
Next, click on Lora
And from here you select the LoRA you want. For this example I used “cns_dress_style3_v2”. If you don’t see your LoRA you can press refresh.
If your LoRA still doesn’t show up you have likely moved the LoRA to the wrong folder.
After you have clicked on your desired LoRA model, new text will appear in the prompt. It will look like this:
The number after the colon (:) indicates the strength of the LoRA, this usually ranges from 0 to 1, however each LoRA is different and I recommend checking the desired settings of your LoRA on the Civit AI webpage where you downloaded your LoRA from, too high values can mess with the image generation.
Now you’re all set to use your new LoRA. Simply click on generate to begin!
To further personalize and enhance the experience, we can associate an image with the LoRA model. By adding a preview image to the model, we can easily identify it in the user interface.
Simply click replace preview after you have generated an image with the LoRA.
LoRA models offer incredible creative possibilities, allowing us to generate images with specific styles, themes, and even cultural elements. By leveraging LoRA models, we can elevate our image creation process and produce stunning results. Thank you for joining us on this journey, and we hope you continue to explore and experiment with LoRA models in your Stable Diffusion endeavors. Happy creating!