How to Use LoRA in Kaiber AI

In this tutorial, we explain how to use LoRA (Low-Rank Adaptation) in Kaiber AI’s Superstudio, along with ModelMaker. This feature allows you to train custom AI models (LoRAs) tailored to your unique style, characters, or creative vision.

Here’s a step-by-step guide:

Step-by-Step: Using LoRA in Kaiber AI

1. Prepare Your Training Data

Choose the Model Type: Decide if you want to train a model for a Character, Place, Style, or Object.
Gather Images: Collect images for training. Recommendations:
- Character/Style: 25–40 images
- Place: 5–25 images
- Object: At least 5 high-quality images
Image Requirements: Use JPG or PNG, max 20MB per image. Ensure images are clear, well-lit, and in focus. Avoid blurry or poor-quality shots.
Add to Collection: Upload your images to your Canvas and group them into a Collection. You can reorder and name your Collection for organization2.

2. Train Your Model

Open Explore Menu: Go to the Explore menu and click “Co-Create.”
Add ModelMaker: Drag the ModelMaker module onto your Canvas.
Connect Collection: Create a new Collection or drag in an existing one, then connect it to the ModelMaker Flow.
Select Training Type: Choose the appropriate type (Character, Place, Style, or Object).
Enter Keyword: Assign a unique keyword for your model.
Generate: Click “Generate” and wait for the process to complete (typically a few minutes)2.

3. Generate with Your Model

Add Image Lab 2.0: Place the Image Lab 2.0 module on your Canvas.
Open Custom Models Menu: In Image Lab 2.0, access the Custom Models menu.
Select Your LoRA: Choose the trained model(s) you want to use (up to 3 different models at once).
Prompt with Keyword: Write a prompt in the Subject Element that includes your model’s keyword. Optionally, add a subject reference image.
Adjust Settings: Use the weights slider if you have multiple models connected, to control their influence on the output.
Generate: Click to generate and explore your results. You can organize outputs in new Collections2.

Tips and Best Practices

Credits Usage: Training a LoRA in ModelMaker costs credits. For example, in Kaiber, it might cost 200–250 credits per LoRA, depending on your plan.
Iterative Work: Many creators use LoRA training for iterative projects, refining their models with each new set of images or outputs.
Prompt Engineering: For best results, craft detailed prompts that reference your LoRA’s keyword and style, especially if you enjoy prompt engineering for creative projects.

Process Summary

Step	Action
Prepare Data	Gather images, upload, and group into a Collection
Train Model	Use ModelMaker, select type, enter keyword, and generate
Generate with LoRA	Use Image Lab 2.0, select custom model, prompt with keyword, generate

This process enables you to create highly personalized AI-generated images and videos in Kaiber AI, leveraging the power of LoRA for fine-tuning and customization.

Related Posts:

Citations:

Step-by-Step: Using LoRA in Kaiber AI

1. Prepare Your Training Data

2. Train Your Model

3. Generate with Your Model

Tips and Best Practices

Footer

OPTIWEB DESIGN