Part 2 of our 3 part Series of articles exploring LoRA Character Bases
If you haven't see the list of Character Base models, check out the previous entry in this series here:
Part 1: Base + Style
Step 1
Pick your first character. For our example we will start with Trista again as a Composition Base.
Step 2
Pick your second character. We will show a few different examples by selecting Gwendolyn, Rochelle, and the Lizbeth models
For the first blend we are going to test a few different prompts with Rochelle.
I'll play with things like hair color, style, and features. Here are some prompts:
teal hair, photography style, dslr, photoshoot
space pilot with minimalist armor
futuristic witch, hyperrealism, photorealistic, ray tracing
Next, we will try a mix of Trista and Gwendolyn
Here are a few more prompts we used with success:
red hair and wearing a pink dress, portrait, photography style, photoshoot, dslr blonde
dark hair, maiden
happy
As you can see with this round, we can get super specific based on what prompts we do or don't use to describe the character and also the strength you set the individual LoRAs to makes a huge difference.
Lastly, to show you what happens when I focus on generating one character but in different poses using two LoRAs, I will pick Lizbeth and Trista.
In this case I am keeping my settings and descriptors completely the same. However, I am going to add different emotions and poses to the prompts:
happy
half torso blue hoodie, photorealistic, white hair, photoshoot, dslr
concerned, photorealistic, white hair, photoshoot, dslr
As you can see there is a lot of potential range, and everything you need to create a dataset to fully train your character.