Type | |
Stats | 203 18 |
Reviews | (31) |
Published | May 20, 2024 |
Base Model | |
Training | Steps: 13,320 Epochs: 90 |
Usage Tips | Clip Skip: 2 |
Trigger Words | ohwx |
Hash | AutoV2 FBB6362F25 |
I made this model with the intention of using it for animations in ComfyUI. The dataset consists of humanoid characters, placed in 2 categories:
1. Fusion of man and machine
2. Fusion of man and nature.
What makes it suitable for animations is the fact that almost 70% of the characters from the training dataset have a simple background behind them (mostly black or white), which I tagged. Using those tags in the prompt will create an image that is very easy to mask, giving you full control over the character and background.
IMPORTANT:
- CLIP SKIP 2
- 80% of the dataset images are: width 512 height 768
- Don't forget to use trigger word, which is: ohwx. The reason I used the trigger word is because it allows us to easily scale the lora weights in the prompt scheduler, for example.
As always, I'm awesome and I'm giving you the full list of captions I used to train the model.
https://drive.google.com/file/d/1kJy763RQEWCA9nRbZzkBi7muOh7bMI4f/view?usp=sharing