How to use Loras with MyAiGF

(This is an advanced tutorial. We recommend completing our easier guides before diving in)

Problem:

You’re able to create the girls you want, but they look like different characters from one scene to another.

We can definitely help, all the tools are here.  We’ll teach you the prompting skills you need and also arm you with a few different character-making strategies.

Solution #1: Use LoRAs to influence your character’s looks

Imagine you’re talking to a friend and you say “she looks like a brunette with black hair”…  this could mean so many different things, right?  Now, if you said “she looks a little like Gal Gadot”, our imagination becomes much clearer. Our system has 1100+ mini-models built specifically for this, whether it is a likeness of a celebrity or a shortcut to getting someone in certain clothes or in a pose.

 

To unlock this super power:

  • Use our advanced weights system.  You can make up a character that is a 20% likeness of Jenny Ortega and 25% and 25% Loren Gray, for example, and that character will
  • Reinforce by adding other celebrity names in the positive prompt box.  The advanced weights system also lets you apply percentages to words without loading a mini-model.
  • Add pose or clothing LoRA to create any scene or situation in a consistent way
  • Use Inpainting when there is more than 1 character in a scene, to avoid traits from jumping to both characters.  Inpainting allows you to mask only the specific section you want to change.

Solution #2: Fine-tune with prompt engineering

You can get even more specific results doing these three things:

  1. Use very specific language to describe them at the beginning of the prompt, such as age and nationality. Each AI model is full of many different characters, so the more specific the description is, the less likely the AI will jump to another character.  Example:  a 23-year-old Korean college student
  2. Make up a number and add it to your prompt, aka a “Seed” number.  A seed is a made-up reference point for prompting. For example, “12345” or “3948394” is a valid seed. They don’t mean anything per se — it doesn’t refer to a specific person or look, but  instead helps the AI understand that you’re talking about the same topic across different prompts.  The command to add a seed is forward slash seed with a colon, like this:   /seed:342342
  3. Stick to one base model and one aspect ratio when possible.  Changing the base art style and dimensions of the image means the AI has to figure out how to fill up the entire image from scratch, so it will result in a totally different image.  Maintaining a text file or spreadsheet with your notes is recommended when creating a persistent character, or use our Archive and Prompt History tools to track back.

Last resort nuclear option: Build your own LoRA

Using a persistent character without long prompts can be a challenge, and it is the main reason why LoRA technology exists.  You can use a website like Dreamlook.ai to upload your generated images and train your own LoRA.

If you are using real photographs to train your model, we only ask that you sign that you have consent to use this character, as deepfakes carry steep fines in some areas. If you’re making a commercial product we strongly advise not using 100% likeness of any living persons. Besides, the AI is hotter in every possible way!