SDXL Dreambooth LORA Training Guide on Google Colab - Unlimited AI Images of Yourself

---

---

AI Summary

Summary: Creating a Low Rank Adaptation (LORA) File for Stable Diffusion Using Google Colab

• Introduction:
  • Previous video: How to create a LORA file for Stable Diffusion.
  • LORA files allow creation of AI images of any person, place, or thing.
  • Issue: High-end PC or MacBook Pro needed.
  • Solution: Free method using Google Colab.
• Steps to Create LORA File:
  1. Go to GitHub, find “autotrain Advanced” repository.
  2. Open in Google Colab via the provided link.
  3. Enter Hugging Face and Enro tokens:
     • Sign up for free accounts if needed.
     • Generate and copy tokens into Colab.
  4. Connect to a GPU instance in Google Colab.
  5. Run the Auto Train code.
  6. Set up Auto Train public URL using Enro.
  7. Access the GUI through the Enro link.
  8. Configure settings:
     • Change task to “Dream Booth LORA”.
     • Select base model (e.g., Stable Diffusion XL).
     • Adjust training parameters.
  9. Upload training files:
     • Use high-quality images from Google Image search.
     • Aim for 5-10 images.
  10. Set prompt and upload images.
  11. Start training:
  • Ignore the warning about charges.
  • Training takes 1-2 hours.
  12. Download the .safe tensor file upon completion.
• Using the LORA File:
  1. Copy the .safe tensor file to the local machine.
  2. Load the LORA file into a Stable Diffusion software (e.g., Focus, invoke AI).
  3. Generate images using the trained model.
• Tips:
  • Use a celebrity lookalike for better training results.
  • Adjust the weight of the LORA to vary adherence to the trained model.
  • Experiment with different art styles and settings in the software.
• Conclusion:
  • The process allows for the creation of personalized AI images.
  • Users can experiment with various styles and settings to produce unique results.
  • The method is accessible for those without high-end hardware.