How to Train Your Own LoRA Model: Complete Guide to Creating AI Influencers

This comprehensive guide will walk you through the entire LoRA training process — from creating your dataset to testing your finished model on druidcat.com.

Training Plan Overview

Here’s what we’ll cover in this guide:

1. Create a Diverse Dataset — Using Qwen Edit and Nano Banana Pro to generate 20-40 images
2. Generate Captions — Using Florence script, ChatGPT, or Gemini Antigravity
3. Launch Runpod Training — Using the Ostris AI Tools template with optimal settings
4. Test Your Results — Upload and use your LoRA on druidcat.com

Step 1: Creating Your Dataset

We begin by creating the face of your AI influencer. Previously, I suggested creating a full body first, but now with Nano Banana Pro generating images in 4K resolution, creating a full body from a face is no longer a problem. Close-up shots capture more facial details, which is essential for consistent LoRA training.

Generating the Initial Face

Go to Kitty AI Studio and select Z-Image Turbo Text to Image. I enjoy the aesthetic of this model as an initial image, but you can also use:

• Instagirl — For influencer-style aesthetics
• WAN Realism — For photorealistic results
• Qwen Text to Image — For natural beauty

Write a detailed prompt. I specify “close up” to get a face shot, then describe the model’s features. For longer hair, consider a medium shot instead. ChatGPT, Grok, or any language model can help you craft better prompts.

Creating the Full Body

Once we have our face, use Nano Banana Pro (Google’s Imagen model, available on Kitty AI Studio) to create the rest of the body. Generate at 4000 pixels to preserve maximum facial details.

Generating Multiple Angles

Upload your full body image to the Qwen Character Angles workflow on the website. After processing, you’ll have photos of your character from different angles.

Diversifying Your Dataset

Now we need to diversify these images. Use these tools to create variations:

• Qwen Edit — Change specific elements while preserving identity
• Qwen Inpaint — Modify specific areas of the image
• Qwen Change Clothes — Swap outfits easily
• Nano Banana Pro — Generate entirely new scenarios

For full body shots, use at least 2000 pixel resolution to preserve facial details. Create variations with:

• Different clothing and outfits
• Different angles and poses
• Different expressions
• Different scenarios (forest, café, studio, etc.)

About 16 images will be enough for the full body portion.

Adding Face Close-ups

Return to face shots. In Nano Banana Pro, select 1000 pixel quality (cheaper and faster). For close-up shots, the face will always have enough detail. Prompt different:

• Expressions (smiling, serious, laughing)
• Angles (front, 3/4 view, profile)
• Blouses or accessories
• Hairstyles

Step 2: Creating Captions

Captions are text files that describe what’s in each image. They are essential for training because they help the AI better understand what it’s learning and respond correctly to prompts. Quality captions significantly improve your LoRA’s results.

Option A: Florence Script (Windows)

I share a Florence script on Patreon that works on Windows. Here’s how to install it:

1. Install Python from python.org. During installation, check “Add Python to Path” — this is critical!
2. Unpack the Florence folder anywhere on your computer
3. Open terminal in the folder: Click the address bar in File Explorer, delete the path, type cmd, and press Enter
4. Install dependencies: pip install -r requirements.txt
5. Run the app: python app.py
6. Open the link shown in terminal in your browser

In the interface:

• Drag and drop your images onto the canvas
• Enter your trigger word in Caption Prefix (e.g., “olivia1”)
• Choose model: Base (faster) or Large (more accurate)
• Click Submit and download the zip with caption files

Option B: ChatGPT (Paid Subscription)

Upload your images to ChatGPT and ask it to create caption text files with the same names as your image files. Request that it adds your trigger word as a prefix to all captions.

Option C: Gemini Antigravity

Download Antigravity from Google, open it in your images folder, and use the same prompt as ChatGPT. Gemini will create the text files directly on your computer.

Step 3: Training on Runpod

Now we need to rent a cloud GPU for a few hours to start training. Go to Runpod and follow these steps:

Setting Up Your Pod

1. Click Pods
2. Select Community Cloud
3. Choose a powerful GPU (RTX 5090 or RTX 6000 recommended)
4. In “Change Templates”, search for “ostris”
5. Select AI Tools template
6. If training multiple models, click Edit Template and set container volume to 250 GB
7. Click Deploy

Connecting to Your Pod

1. When the pod finishes loading, click Connect
2. Select port 8675
3. Enter password: password (default for this template)
4. Navigate to Datasets and upload all your files (images + text files in one folder)

Configuring Z-Image Training

Click on the New Job tab and configure:

In the Sample tab:

• Remove all example prompt boxes except one
• Paste a prompt from your captions but place your influencer in a new scenario (e.g., “sitting in a café”)
• These control images help monitor training progress

Click Create Job, then click the Play button (triangle) in the Dashboard.

Settings for WAN 2.2 (14B)

The main differences for WAN 2.2 training:

• This produces TWO files: High Noise and Low Noise — download both!
• In Sample tab: Set Number of Frames = 1 and FPS = 1
• Enable Skip First Sample (we’re training for images, not video)
• Everything else is the same as Z-Image

After a few hours, your LoRA models will be ready for download in the Dashboard under Checkpoints.

Step 4: Testing Your LoRA

Go back to Kitty AI Studio to test your newly trained LoRA.

Testing Z-Image LoRA

1. Go to Z-Image Text to Image workflow
2. Enter your prompt including your trigger word
3. Check the LoRA box and upload your file
4. Click Generate

Testing WAN LoRA

1. Go to WAN 2.2 Text to Image
2. Add BOTH the High Noise and Low Noise LoRA files
3. Include your trigger word in the prompt
4. Click Generate

Conclusion

You now have a complete toolkit for creating AI influencers. You can monetize this content on various platforms and social media. The key is consistency — your trained LoRA ensures your AI character looks the same across all generations.