Welcome to part two of our Magai Image Editor series, where we'll explore how to create and use LoRAs. A LoRA (Low-Rank Adaptation) is essentially a custom-trained image model that can produce any subject of your choice in different scenarios. One of the most popular use cases is creating digital photoshoots of yourself or your team—without ever stepping into a studio.
This guide will walk you through the complete process of preparing your images, creating a LoRA, and using it effectively within Magai.
Preparation: Getting Your Images Ready
Step 1: Collect Your Images
Start by gathering at least 10 good photos of your subject. For the best results:
Include as much variety as possible
Use different backgrounds
Include multiple poses and expressions
Capture different angles
Include shots from various distances
The more variety you provide, the more versatile your LoRA will be. While 10 images is the minimum, more images will generally yield better results.
Step 2: Convert Image Formats
Ensure all your photos are in JPEG or PNG format. iPhones often save images in HEIC format, which isn't compatible with the LoRA training process.
For Mac users:
Right-click or Command-click on the image
Select "Quick Actions"
Click "Convert Image"
Select JPEG format
Click "Convert"
For PC users: Similar conversion tools are available through the File Explorer or third-party applications.
Step 3: Create a Zip File
Compress all your properly formatted images into a single zip file:
For Mac users:
Select all the images
Right-click and select "Compress"
For PC users:
Select all the images
Right-click and select "Send to" > "Compressed (zipped) folder"
Creating Your LoRA in Magai
Step 1: Access the Image Editor
Go to the Magai dashboard
Click on "Quick Actions"
Select "Create a new image"
In the image editor, select "Flux LoRA" from the model list
Step 2: Configure Your LoRA
The first time you create a LoRA, you'll see a blue button explaining the process:
Initial LoRA setup consumes 11,000 words
You'll need a prepared zip file with your images
Once trained, generating images with your LoRA costs the same as any other image
Click "Setup Model" and configure the following:
1. Name Your LoRA
Choose a name that helps you remember what the model is trained on. Consider including settings information in the name, for example:
"Dustin_P_M_4K" (where P = portrait mode, M = mask, 4K = 4000 steps)
2. Set a Trigger Word
The trigger word is what you'll type in your prompts to activate your LoRA:
For a person, use their name (e.g., "dustin")
For common names, add specificity (e.g., "dustin001") to avoid confusion with existing entities in the AI's training
For non-human subjects like products, consider using a unique identifier
3. Select Training Options
Portrait Model:
Enable this if you're creating photorealistic portraits of people
Optimizes for realistic portrait generation with bright highlights and detailed results
Style:
Use this if you're training the model to replicate a style rather than a subject
Good for photographers wanting to recreate their aesthetic
Create Mask:
Tells the AI to focus on the subject and ignore backgrounds
Recommended for most portrait-based LoRAs
Can still be used without portrait mode if you want non-photorealistic renditions
Multi-resolution Training:
Enable this if your training photos have varying resolutions
Helps achieve better results with mixed-quality source images
4. Set Number of Steps
The number of steps determines how strictly the AI will recreate your subject:
Lower steps (1000-3000): More versatile, can place your subject in different scenarios but may occasionally be less accurate
Higher steps (4000+): More accurate resemblance but less creative freedom and versatility
Recommendation: 2000 steps works well for most purposes, balancing accuracy with versatility.
5. Upload Your Zip File
Drop your prepared zip file into the uploader and wait for it to complete.
Step 3: Create Your LoRA
Click "Create LoRA" to begin the training process. This typically takes about five minutes.
Important note: Once a LoRA is trained, it cannot be modified. If you want to make changes, you'll need to train a completely new model.
Using Your LoRA
Once training is complete, your new LoRA will appear in the dropdown menu when using the Flux LoRA model. To generate images with your custom LoRA:
Select Flux LoRA as your model
Choose your custom LoRA from the dropdown
Include your trigger word in your prompt
Add any additional descriptions for poses, settings, or styles
For example: "dustin wearing a business suit in a modern office, professional portrait, high quality"
Tips for Best Results
Be specific with your prompts
Include your trigger word near the beginning of your prompt
Experiment with different styles and scenarios
Use negative prompts to avoid unwanted elements
Try different step counts to find the sweet spot for your specific LoRA
Conclusion
Creating custom LoRAs in Magai opens up endless creative possibilities for personalized AI-generated imagery. Whether you're creating professional headshots, concept art, or creative portraits, your custom LoRA will help you achieve consistent, personalized results that accurately represent your subject.
Remember that the quality of your input images greatly affects the quality of your LoRA, so take time to select diverse, high-quality images for the best results.## Introduction