🍁Welcome to Henesys the town of Floral Life 🍁
Description
Trained on: 25 images
Training Model: Animefull-final-pruned-fp16
flip_aug: False
Num of Repeats: 10
Unit is Epochs or Steps: Epochs
Number of Epochs or Steps: 10
Training Batch Size: 2
Total Steps: 1250
Resolution: 512
Network Dim: Dim 16 and Alpha 8
Lora Creation Process took: 00:12:59
FAQ
Comments (4)
Could you please provide more details about the training? Thank you.
Yes. I just updated the "About this version" menu on the sidebar on the right side of the page.
From there you'll be able to see all my training settings.
Let me know if you have any other questions :)
@CitronLegacy Thanks! I would like to ask if there are any tips for choosing the training set and tagging? Additionally, what range of loss values should I aim for to achieve good results?
@e24106204175
Here is a preview of what the training data looked like: https://civitai.com/images/20432384
Basically you should images for each concept/idea/thing/person/etc you are trying to make. (Characters are easier to make so let me know if you tips specifically for characters). For buildings its important to have pictures of the same or similar buildings with slight adjustments so the AI has some creativity about how to make variations. (For example a mushroom house with a orange roof and a house with a yellow roof will help the AI to understand changing roof colors)
As for Loss values: I always use defaults in the trainer (Either Holostrawbery's Google Colab or CivitAI's trainer) and never think about it.
This is the guide I used when I first got started https://civitai.com/models/22530?modelVersionId=642321
This guide will help with general advice for picking images as well as choosing training settings.