Integrating LoRa Character Bases in Game Design Part 1: Base + Style

We just made creating your next Consistent Character much easier

On January 29th, 2024 we introduced a collection of LoRA's which act as 'Character Bases'. Here is a list of the Model Names of our initial release:

You can use these Character Bases in the following ways:

• On their own as a consistent character
• As a base to create a new consistent character
• Build a new consistent character dataset

All of these can be found in Community Models and the LoRA Composition Library. You can also search them by name in either library.

We call these LoRAs "Character Bases" because they have been trained in such a way that they trend towards character consistency by default and have been designed to look like a particular character.

Character bases are also SUPER FLEXIBLE, and easily be reworked towards different character designs.

There are three potential basic formulas for creating a NEW consistent character from a Character Base.

Click on the links above to check them all out. This one is the first Character Base + Style workflow.

First we will look at one of these LoRAs on it's own. We're pleased to introduce Trista, a really flexible Character Base model which defaults to a realistic brunette woman.

By default, Trista is very easy to prompt. You can simply guide the model with such prompts as "wearing a beret" or "laughing"

You can also start to push her in the direction of a different character design by describing features, like her hair color, eye color, fashion sense, etc.

Here we have just prompted "red hair"

You can build a totally new dataset and character dataset in this way simply with prompts if you choose.

But let's take it a step further!

We'll first show you how you can create a consistent, super stylized character by using Trista as a base. You'll need to click "New Model" and "Start Composing".

The beauty of a Character Base is that its strongest attribute is consistency.

In fact, you can completely change the style of your character if you wish. Here we are going to blend together Trista with Blocky Cartoons.

To guide the output you simply adjust the settings.

If I want the end result to be more "cartoonish" you should give more strength to the Cartoon LoRA and less strength to Trista.

Or you can also simply add "cartoon" to the prompt to give that underlying style a little extra oomph.

Because this cartoon style has a lot of different poses and views of characters, it's really easy to guide it with prompts like "yelling," "happy," "portrait," and "full body".

We also included "dark hair" in the prompt because we know Trista's model is a little flexible when it comes to hair color. If there are details you want to lock in that aren't remaining consistent, just prompt them!

You can also go a little heavier with the cartoon style and break out of the basic look that Trista has. In this case, we can prompt different hair colors or descriptive features.

The amazing thing about the Character Base is the Consistency. Even when we're moving away from the basic description of the original character, that idea of "consistency" remains in the model.

If you need it to be slightly more consistent, those results are usually 1 or 2 descriptive prompts away: hair color, eye color, outfit description, etc

These images were made with the same model using slightly higher Blocky Cartoon influence and the basic "portrait" prompt.

All of this is a good way to start building a dataset for your new character. You don't really need to fight with other AI to get exact consistency. If you find a character you really like and want to lock it in, refer to our basic guide to curating and captioning a dataset here:

Basic Tips for Curating & Captioning a Dataset: A Beginner's Guide

And be sure to check out Part 2:

Integrating LoRa Character Bases in Game Design Part 2: Base + Base