Adapted to Google Colab based on Kohya Guide
Adapted to Google Colab by Linaqruf
You can find latest notebook update here

1. Using the U-Net learning
2. Automatic captioning/tagging for every image automatically with BLIP/DeepDanbooru
3. Implemented NovelAI Aspect Ratio Bucketing Tool so you don't need to crop image dataset 512x512 ever again

• Use the output of the second-to-last layer of CLIP (Text Encoder) instead of the last layer.
• Learning at non-square resolutions (Aspect Ratio Bucketing) .
• Extend token length from 75 to 225.

4. By preparing a certain number of images (several hundred or more seems to be desirable), you can make learning even more flexible than with DreamBooth.
5. It also support Hypernetwork learning
6. NEW! 23/11 - Implemented Waifu Diffusion 1.4 Tagger for alternative DeepDanbooru to auto-tagging

• Install gallery-dl to scrap images, so you can get your own dataset fast with google bandwidth
• Huggingface Integration, here you can login to huggingface-hub and upload your trained model/dataset to huggingface

Kohya | Just for my part