In this quick post, I'm going to share my worfkflow to create a LoRA with a single picture. If you've been following my work from the very start, you'll know that my first LoRAs were mainly traine on single pictures (see my 1-shot series on CivitAI).
The idea of this post is not to explain in detail each parameter available, but to give you a quick setup to train your own LoRAs.

The UI I'm going to use for the training is Kohya_ss, which is one of the most used at the moment.
I'm not going to do a step-by-step setup tutorial in this post (you can find a lot of tutorials about it just by searching on Bing), but you can learn more about how to install it here:
bmaltais/kohya_ss (github.com)
After you've installed the UI just launch it using the commands explained here:
bmaltais/kohya_ss (github.com)

The first step is (obviously) to get a picture to train the LoRA on. In my workflow I try to avoid training LoRAs on non-public domain content, so for this example I'm going to use Bing.com/create (but you can use what you prefer).
After we've opened the page, we can now use a prompt. In the below examples I've seen one that it seems interesting:

Let's change its prompt to fit what I want to create:

woman portrait, sufganiot, byzantine Japanese woodblock scene, in viennese secession style, gilded, chromatic, whimsical, hanukkah
Copy

And here's the batch of pictures:

From this batch I'd like to train a LoRA based on the picture on the bottom right. Keep in mind that when training a LoRA on a single picture, it will be highly biased towards each characteristic of the picture (so in this case it will probably be biased towards candelabras and hair decorations).

To train the model you'll need to create a folder with a specific structure:

| model-name
| -- img
| ---- pic-folders
| -- model
Copy

• model-name: This can be anything you like. For this example I'm going to call it learn_final (because for some reason I've called all my project folders something_final)
• img: This is the folder where you'll put the pictures subfolders
• pic-folders: This can be one or more folders (in our simple case it will be just one). The name of this folder should be the number of the repetitions, underscore, a list of tags divided by commas. For this example I'm going to use 40 repetitions with just an "inzaniak" tag (so the folder will be 40_inzaniak). If I'm not training a specific concept I prefer to keep just the "inzaniak" tag without adding anything else. If you want, you can also add a .txt file in the same folder (keeping the same name of the picture), with a list of specific tags related to the picture.
• model: This is the folder, where the model will be saved
  So in my case this is the actual folder structure:

\learn_final
├───img
│   └───40_inzaniak
│           OIG.jpg
│
├───log
└───model
Copy

Remember to copy the picture you've downloaded in the first step to the picture folder!

The first thing we need to do is to go to the LoRA tab:

Once you are in the tab we need to do the following:

• Select the source model
• Select the folders we've created
• Choose the parameters for the training

Bonus/Super patrons can find the pre-configured train file here:

Once you've finished training the LoRA, you should now see the trained models inside of the model folder. You can copy every .safetensors file and paste it in your LoRA folder.
Now you just need to run a prompt as you usually do and experiment with the different epochs.
Here's a bunch of pictures generated with Ranbooru with the 8 epochs version of the LoRA and a weight of 1.0:

I've started developing custom models for myself a few months ago just to check out how SD worked, but in the last few months it has become a new hobby I like to practice in my free time. All my checkpoints and LoRAs will always be released for free on Patreon or CivitAI, but if you want to support my work and get early access to all my models feel free to check out my Patreon:
https://www.patreon.com/Inzaniak

If you want to support my work for free, you can also check out my music/art here:

• Bandcamp
• DeviantArt